U.S. patent application number 13/000602 was filed with the patent office on 2011-05-05 for controller.
This patent application is currently assigned to HONDA MOTOR CO. LTD.. Invention is credited to Michael Fischer, Yuji Yasui.
Application Number | 20110106402 13/000602 |
Document ID | / |
Family ID | 41506745 |
Filed Date | 2011-05-05 |
United States Patent
Application |
20110106402 |
Kind Code |
A1 |
Yasui; Yuji ; et
al. |
May 5, 2011 |
CONTROLLER
Abstract
A controller capable of immediately and high accurately
controlling a detected value to a desired value in a controlled
object having a large delay characteristic is provided. The
controller for controlling a plant comprises a predictor (511) for
calculating a predicted value (PREDNTH3.sub.EXS) of the future of a
control output value based on a provisional value
(DG.sub.UREA.sub.--.sub.EXS) of a control input value including a
periodic reference signal by using a plant model indicating the
dynamic characteristics of the control output value from the
control input value of the plant, an evaluation function value
calculator (512) for calculating an evaluation function value (J)
including the predicted value (PREDNH3.sub.EXS) of the future of
the calculated control output value, an extreme value search
optimizer (513) for calculating a provisional value
(DG.sub.UREA.sub.--.sub.OPT) of such a control input value as that
the evaluation function value (J) becomes the extreme value on the
basis of a product (C.sub.R) of the calculated evaluation function
value (J) and a periodic reference signal (S.sub.REF), and an adder
(54) for calculating a control input value (G.sub.UREA) including
the provisional value (DG.sub.UREA.sub.--.sub.OPT) of the
calculated control input value.
Inventors: |
Yasui; Yuji; (Saitama,
JP) ; Fischer; Michael; (Offenbach/Am Main,
DE) |
Assignee: |
HONDA MOTOR CO. LTD.
TOKYO
JP
|
Family ID: |
41506745 |
Appl. No.: |
13/000602 |
Filed: |
July 7, 2008 |
PCT Filed: |
July 7, 2008 |
PCT NO: |
PCT/JP2008/062291 |
371 Date: |
December 21, 2010 |
Current U.S.
Class: |
701/102 |
Current CPC
Class: |
F01N 2900/10 20130101;
Y02A 50/20 20180101; F01N 2900/1402 20130101; F01N 2610/02
20130101; Y02T 10/12 20130101; F01N 2560/026 20130101; Y02T 10/24
20130101; Y02A 50/2325 20180101; F01N 2900/0408 20130101; F01N
13/0093 20140601; F01N 2900/1404 20130101; G05B 13/048 20130101;
F01N 3/103 20130101; F01N 2900/0411 20130101; F01N 3/208 20130101;
F01N 2560/06 20130101; F01N 13/0097 20140603; F01N 2560/021
20130101 |
Class at
Publication: |
701/102 |
International
Class: |
F01N 9/00 20060101
F01N009/00 |
Claims
1. A controller for controlling a plant, comprising: a predicted
value calculating means for calculating, using a plant model
showing a dynamic characteristic of a control output value from a
control input value of the plant, a future predicted value of a
control output value based on a provisional value of a control
input value containing a periodic reference signal; an evaluation
function value calculating means for calculating an evaluation
function value containing the future predicted value of the control
output value thus calculated; an extremum searching means for
calculating, based on a product of the evaluation function value
thus calculated and the periodic reference signal, a provisional
value of a control input value so that the evaluation function
value becomes an extremum; and a control input value calculating
means for calculating a control input value containing the
provisional value of the control input value thus calculated.
2. The controller according to claim 1, wherein an update period of
a provisional value of a control input value is shorter than an
update period of the control input value.
3. The controller according to claim 1, further comprising a target
value correction means for setting, for a target value of a control
output value, a corrected target value of the control output value
between the target value of the control output value and the
control output value, wherein the evaluation function value
calculating means calculates the evaluation function value based on
a deviation between the corrected target value of the control
output value thus calculated and the future predicted value of the
control output value thus calculated.
4. The controller according to claim 1, wherein the predicted value
calculating means calculates future predicted values of a control
output value at a plurality of different times, and wherein the
evaluation function value calculating means calculates an
evaluation function value containing the future predicted values of
the control output value at the plurality of different times thus
calculated.
5. The controller according to claim 4, wherein the plant model
showing the dynamic characteristic of the control output value from
the control input value of the plant includes a plurality of
control input terms proportional to the control input value, and a
plurality of control output terms proportional to the control
output value, and wherein the predicted value calculating means
calculates a future predicted value of the control output value at
a plurality of different times by recursively using output of the
plant model in the control output terms.
6. The controller according to claim 5, wherein the predicted value
calculating means calculates a future predicted value of a control
output value at a plurality of different times by recursively
using, in the control output terms, output of the plant model in
which a provisional value of a control input value not containing a
periodic reference signal was used in the control input terms.
7. The controller according to claim 5, wherein the predicted value
calculating means uses a provisional value of a control input value
containing a periodic reference signal only in a portion of the
plurality of control input terms.
8. The controller according to claim 1, wherein the plant is an
exhaust purification device including: a selective reduction
catalyst that is provided in an exhaust channel of an internal
combustion engine, and reduces NOx flowing through the exhaust
channel under the presence of a reducing agent; a reducing agent
supply means for supplying a reducing agent into the exhaust
channel on an upstream side of the selective reduction catalyst;
and an exhaust detection means for detecting exhaust in the exhaust
channel on a downstream side of the selective reduction catalyst,
wherein control output values of the plant include an output value
of the exhaust detection means, and wherein control input values of
the plant include a supply amount of the reducing agent from the
reducing agent supply means.
9. The controller according to claim 1, wherein the plant is a
vehicle including: a wheel; a drive source that produces torque for
rotationally driving the wheel; and a drive wheel speed detection
means for detecting a rotational speed of the wheel that is
rotationally driven by the drive source, wherein control output
values of the plant include a detected value of the drive wheel
speed detection means, and wherein control input values of the
plant include a target value of torque of the drive source.
Description
TECHNICAL FIELD
[0001] The present invention relates to a controller. Specifically,
it relates to a controller of a plant having a large response delay
characteristic.
BACKGROUND ART
[0002] Conventionally, as one exhaust purification device that
purifies NOx in exhaust, a device has been proposed in which a
selective reduction catalyst that selectively reduces NOx in the
exhaust by adding a reducing agent is provided in an exhaust
channel. For example, a selective reduction catalyst of urea
addition type that uses urea water as a reducing agent generates
ammonia from the urea thus added, and selectively reduces NOx in
the exhaust with this ammonia.
[0003] With such a selective reduction catalyst, in a case in which
the injection amount of reducing agent is less than the optimum
amount, the NOx reduction rate declines from the ammonia consumed
in the reduction of NOx being deficient, and in a case of being
greater than this optimum amount, the ammonia that is surplus in
the reduction of NOx is discharged. As a result, with exhaust
purification devices equipped with a selective reduction catalyst,
it has been important to appropriately control the injection amount
of the reducing agent. Consequently, devices that estimate the NOx
reduction rate of the selective reduction catalyst and control the
injection amount of the reducing agent based on this estimation are
exemplified in Patent Document 1 and Patent Document 2.
[0004] The exhaust purification device of Patent Document 1 detects
the NOx concentration on a downstream side of the selective
reduction catalyst, and estimates the composition of exhaust
flowing into the selective reduction catalyst, i.e. the ratio of NO
to NO.sub.2, from this detected NOx concentration and the operating
state of the internal combustion engine. Furthermore, it estimates
the NOx reduction rate of the selective reduction catalyst based on
the composition of this exhaust, and controls the injection amount
of the reducing agent.
[0005] In addition, the exhaust purification device of Patent
Document 2 detects the temperature of the catalyst as an amount
related to the NOx reduction rate of the selective reduction
catalyst, and controls the injection amount of the reducing agent
based on this temperature.
[0006] However, the NOx reduction rate of the selective reduction
catalysts change not only by the aforementioned such composition of
the exhaust and temperature of the selective reduction catalyst,
but also according to the degradation state of the selective
reduction catalyst. In addition, there is variation in purification
performance between individual catalysts. In addition to this, in a
case of ammonia being stored in the selective reduction catalyst,
the NOx reduction rate of the selective reduction catalyst will
apparently change due to the optimum amount of reducing agent
differing. Therefore, it is difficult to always optimally control
the injection amount of reducing agent with exhaust purification
devices such as those exemplified in Patent Documents 1 and 2.
[0007] Consequently, technology to more directly detect the NOx
purification rate of a selective reduction catalyst and to control
the injection amount of reducing agent based on this will be
considered hereinafter.
[0008] FIG. 16 is a schematic diagram showing a configuration of a
conventional exhaust purification device 80.
[0009] As shown in FIG. 16, in an exhaust channel 82 of an engine
81 are provided, in order from an upstream side to a downstream
side, an oxidation catalyst 83, a urea injection value 85 that
injects urea water as a reducing agent, which is stored in a urea
tank 84, into the exhaust channel 82, and a selective reduction
catalyst 86 that reduces NOx in the exhaust under the presence of
urea water. In addition, as a device that monitors the purification
performance of the selective reduction catalyst, a temperature
sensor 87 that detects the temperature of the selective reduction
catalyst 86 and a NOx sensor that detects the NOx concentration on
a downstream side of the selective reduction catalyst 86 are
provided.
[0010] With this exhaust purification device 80, for example, the
NOx concentration of exhaust emitted from the engine 81 is
estimated using a map established beforehand, and an injection
amount of urea water from the urea injection value 85 is determined
based on this NOx concentration and the catalyst temperature
detected by the temperature sensor 87. In particular, the
degradation state of the selective reduction catalyst 86 can be
estimated here based on a difference between the NOx concentration
detected by the NOx sensor 88 and the NOx concentration of exhaust
estimated. With this exhaust purification device, it is possible to
correct the injection amount of urea water according to the
degradation state of the selective reduction catalyst 86 estimated
in the above way. [0011] Patent Document 1: Japanese Unexamined
Patent Application Publication No, 2006-274986 [0012] Patent
Document 2: Japanese Unexamined Patent Application Publication No.
2004-100700
DISCLOSURE OF THE INVENTION
Problems to be Solved by the Invention
[0013] FIG. 17 presents graphs showing relationships between a NOx
concentration and ammonia concentration in exhaust downstream of
the selective reduction catalyst and the output value of the NOx
sensor in a conventional exhaust purification device. More
specifically, FIG. 17 shows, in order from the top, relationships
between the urea injection amount and the NOx concentration of
exhaust downstream of the selective reduction catalyst, the ammonia
concentration of exhaust downstream of the selective reduction
catalyst, and the output value of the NOx sensor.
[0014] When the injection amount of urea water increases, the NOx
reduction rate of the selective reduction catalyst rises due to the
ammonia generated in the selective reduction catalyst also
increasing. As a result, the NOx concentration on a downstream side
of the selective reduction catalyst decreases accompanying the
injection amount of urea water increasing, as shown in FIG. 17. In
addition, if the urea water injection amount shown by the star is
exceeded, the NOx concentration becomes substantially constant
irrespective of the urea water injection amount. In other words,
the urea water of an amount exceeding the star indicates being
surplus in reducing the NOx generated.
[0015] In addition, herein, the ammonia generated from the urea
water that becomes surplus is not consumed in the reduction of NOx,
and is stored in the selective reduction catalyst or discharged to
downstream of the selective reduction catalyst. Therefore, as shown
in FIG. 17, the ammonia concentration of exhaust downstream of the
selective reduction catalyst increases when the injection amount of
urea water shown by the star is exceeded. It should be noted that
the generated ammonia not being stored in the selective reduction
catalyst and being discharged to downstream thereof in this way is
hereinafter referred to as "ammonia slip".
[0016] As described above, since the urea water injection amount
indicated by the star in FIG. 17 can both minimize the NOx
concentration and ammonia concentration, it is the optimum
injection amount for this exhaust purification system.
[0017] However, as shown in FIG. 17, the output value of the NOx
sensor shows a downward convex characteristic in which the output
value of this optimum injection amount is a minimum point. This is
because existing NOx sensors in the detection principles thereof
are sensitive not only to NOx, but also to ammonia.
[0018] Therefore, it is impossible to determine whether the
injection amount of urea water is insufficient relative to the
optimum injection amount or excessive, with only the output value
from the NOx sensor. As a result, it is difficult to suppress the
discharge of ammonia while continually supplying urea water of an
optimum amount to maintain the NOx reduction rate of the selective
reduction catalyst to be high.
[0019] Consequently, a case of using an ammonia sensor without
using a NOx sensor will be considered. Although existing NOx
sensors are sensitive not only to NOx, but also to ammonia, it has
been known that is it possible to develop ammonia sensors that are
sensitive to only ammonia without being sensitive to NOx.
[0020] FIG. 18 presents graphs showing relationships between the
NOx concentration and ammonia concentration in exhaust downstream
of the selective reduction catalyst and the output value of the
NH.sub.3 sensor in a conventional exhaust purification device using
an ammonia sensor.
[0021] As shown in FIG. 18, the output value of the ammonia sensor
differs from the output value of the NOx sensor and does not become
downward convex, and shows an output characteristic proportional to
the actual ammonia concentration downstream of the selective
reduction catalyst. As a result, it has been considered to perform
feedback control on the urea water injection amount so as to
suppress ammonia slip based on the output value of the ammonia
sensor, by way of a conventionally known control method such as PID
control or sliding mode control.
[0022] However, it has also been known that the ammonia slip of the
selective reduction catalyst has a large response delay relative to
the injection amount of urea water. In general, in a case of
performing feedback control on a control object having a large
delay characteristic, the tracking characteristic of the detected
value to the target value is poor, and it takes time until the
detected value matches the target value. In addition, if the
tracking characteristic of the detected value to the target value
is improved, overshoot and hunting will occur. In particular, in a
case of the aforementioned such exhaust purification device using a
selective reduction catalyst being set as the control object, if
the detected value cannot be quickly and highly precisely
controlled to the target value, the NOx reduction rate of the
selective reduction catalyst will decline, and ammonia slip will
occur.
[0023] The present invention has been made taking the
aforementioned points into consideration, and has an object of
providing a controller of a control object having a large delay
characteristic that can quickly and highly precisely control a
detected value to a target value.
Means for Solving the Problems
[0024] In order to achieve the above-mentioned object, according to
a first aspect of the invention, a controller (3, 7) for
controlling a plant (2, 6) includes: a predicted value calculating
means (511, 712) for calculating, using a plant model showing a
dynamic characteristic of a control output value (NH3.sub.CONS,
DNH3.sub.CONS, M.sub.S.sub.--.sub.ACT) from a control input value
(G.sub.UREA, TRQ) of the plant, a future predicted value
(PREDNH3.sub.EXS, PREW.sub.S.sub.--.sub.EXS) of a control output
value based on a provisional value (DG.sub.UREA.sub.--.sub.EXS,
DTRQ.sub.EXS, TRQ.sub.EXS) of a control input value containing a
periodic reference signal (S.sub.REF, S.sub.REF'); an evaluation
function value calculating means (512, 713) for calculating an
evaluation function value (J, J', J'') containing the future
predicted value (PREDNH3.sub.EXS, PREW.sub.S.sub.--.sub.EXS) of the
control output value thus calculated; an extremum searching means
(513, 714) for calculating, based on a product (C.sub.R,
C.sub.R.sub.--.sub.AVE, C.sub.R', C.sub.R.sub.--.sub.AVE') of the
evaluation function value (J, J', J'') thus calculated and the
periodic reference signal (S.sub.REF, S.sub.REF') a provisional
value (DG.sub.UREA.sub.--.sub.OPT, DG.sub.UREA.sub.--.sub.EXS,
DTRQ.sub.OPT, DTRQ.sub.EXS) of a control input value so that the
evaluation function value (J, J', J'') becomes an extremum; and a
control input value calculating means (54, 76) for calculating a
control input value (G.sub.UREA, TRQ) containing the provisional
value (DG.sub.UREA.sub.--.sub.OPT, DTRQ.sub.OPT) of the control
input value thus calculated.
[0025] According to a second aspect of the invention, in the
controller (3, 7) as described in the first aspect, an update
period (.DELTA.Te) of a provisional value
(DG.sub.UREA.sub.--.sub.OPT, DG.sub.UREA.sub.--.sub.EXS,
DTRQ.sub.OPT, DTRQ.sub.EXS) of a control input value is shorter
than an update period (.DELTA.Ti, .DELTA.Tc) of the control input
value (G.sub.UREA, TRQ).
[0026] According to a third aspect of the invention, the controller
(3, 7) as described in the first or second aspect further includes:
a target value correction means (52, 74) for setting, for a target
value (NH3.sub.CONS.sub.--.sub.TRGT, W.sub.S.sub.--.sub.CMD) of a
control output value, a corrected target value
(NH3.sub.CONS.sub.--.sub.TRGT.sub.--.sub.MOD,
W.sub.S.sub.--.sub.CMD.sub.--.sub.MOD) of the control output value
between the target value (NH3.sub.CONS.sub.--.sub.TRGT,
W.sub.S.sub.--.sub.CMD) of the control output value and a control
output value (NH3.sub.CONS, W.sub.S.sub.--.sub.ACT), in which the
evaluation function value calculating means calculates the
evaluation function value (J, J', J'') based on a deviation between
the corrected target value
(NH3.sub.CONS.sub.--.sub.TRGT.sub.--.sub.MOD,
W.sub.S.sub.--.sub.CMD.sub.--.sub.MOD) of the control output value
thus calculated and the future predicted value (PREDNH3.sub.EXS,
PREW.sub.S.sub.--.sub.EXS) of the control output value thus
calculated.
[0027] According to a fourth aspect of the invention, in the
controller (3, 7) as described in any one of the first to third
aspects, the predicted value calculating means calculates future
predicted values (PREDNH3.sub.EXS, PREW.sub.S.sub.--.sub.EXS) of a
control output value at a plurality of different times, and the
evaluation function value calculating means calculates an
evaluation function value (J, J', J'') containing the future
predicted values of the control output value at the plurality of
different times thus calculated.
[0028] According to a fifth aspect of the invention, in the
controller (3, 7) as described in the fourth aspect, the plant
model showing the dynamic characteristic of the control output
value (NH3.sub.CONS, DNH3.sub.CONS, W.sub.S.sub.--.sub.ACT) from
the control input value (G.sub.UREA, TRQ) of the plant includes a
plurality of control input terms proportional to the control input
value, and a plurality of control output terms proportional to the
control output value, and the predicted value calculating means
calculates a future predicted value (PREDNH3.sub.EXS,
PREW.sub.S.sub.--.sub.EXS) of the control output value at a
plurality of different times by recursively using output of the
plant model in the control output terms.
[0029] According to a sixth aspect of the invention, in the
controller (3, 7) as described in the fifth aspect, the predicted
value calculating means calculates a future predicted value
(PREDNH3.sub.EXS, PREW.sub.S.sub.--.sub.EXS) of a control output
value at a plurality of different times by recursively using, in
the control output terms, output of the plant model in which a
provisional value (DG.sub.UREA.sub.--.sub.OPT, DTRQ.sub.OPT) of a
control input value not containing a periodic reference signal was
used in the control input terms.
[0030] According to a seventh aspect of the invention, in the
controller (3, 7) as described in the fifth or sixth aspect, the
predicted value calculating means uses a provisional value
(DG.sub.UREA.sub.--.sub.EXS, DTRQ.sub.EXS) of a control input value
containing a periodic reference signal (S.sub.REF, S.sub.REF') only
in a portion of the plurality of control input terms.
[0031] According to an eighth aspect of the invention, in the
controller (3) as described in any one of the first to seventh
aspects, the plant is an exhaust purification device including: a
selective reduction catalyst (231) that is provided in an exhaust
channel of an internal combustion engine, and reduces NOx flowing
through the exhaust channel under the presence of a reducing agent;
a reducing agent supply means (25) for supplying a reducing agent
into the exhaust channel on an upstream side of the selective
reduction catalyst; and an exhaust detection means (26) for
detecting exhaust in the exhaust channel on a downstream side of
the selective reduction catalyst, in which control output values of
the plant include an output value (NH3.sub.CONS) of the exhaust
detection means, and control input values of the plant include a
supply amount (G.sub.UREA) of the reducing agent from the reducing
agent supply means.
[0032] According to a ninth aspect of the invention, in the
controller (7) as described in any one of the first to seventh
aspects, the plant is a vehicle including: a wheel (61); a drive
source that produces torque for rotationally driving the wheel; and
a drive wheel speed detection means for detecting a rotational
speed of the wheel that is rotationally driven by the drive source,
in which control output values of the plant include a detected
value (W.sub.S.sub.--.sub.ACT) of the drive wheel speed detection
means, and control input values of the plant include a target value
(TRQ) of torque of the drive source.
Effects of the Invention
[0033] According to the first aspect of the invention, the future
predicted value of a control output value based on a provisional
value of a control input value containing the periodic reference
signal is calculated using a plant model showing the dynamic
characteristic of the control output value from the control input
value of the plant. Furthermore, the evaluation function value
containing the future predicted value of this control output value
is calculated, a provisional value of the control input value such
that the evaluation function value becomes an extremum is
calculated based on the product of this evaluation function value
and the reference signal, and a control input value containing this
provisional value of the control input value is calculated.
[0034] With this, when controlling a plant having a large response
delay between a control input value and a control output value, it
is possible to cause this control output value to match the target
value in a short time with high precision, while suppressing
overshoot and oscillatory behavior of the control output value
relative to a predetermined target value. In addition, this enables
the operational load to be reduced on the order of several to tens
of times compared to model prediction control by convention
evaluation norm. Therefore, even in a case in which a computing
device having high computing power cannot be used due to being in
an adverse environment such as high temperature, high humidity,
high vibration, or dust, it is possible to perform control with
high precision using an computational device of a simple
configuration.
[0035] According to the second aspect of the present invention, the
update period of this provisional value of the control input value
is set to be shorter than the update period of the control input
value actually input to the plant. This enables oscillations when
optimizing the provisional value of the control input value to be
prevented from appearing in the control input value input to the
plant. Therefore, it is possible to stably control the plant.
[0036] According to the third aspect of the present invention the
corrected target value, for the target value of the control output
value, is set between this target value and the control output
value, the evaluation function value is calculated based on the
deviation between this corrected target value and the future
predicted value of the control output value, and the provisional
value of the control input value is calculated based on this
evaluation function value. This enables overshoot and oscillatory
behavior of the control output value relative to the target value
to be further suppressed.
[0037] According to the fourth aspect of the present invention,
future predicted values of the control output value are calculated
at a plurality of different times, an evaluation function value
containing these future predicted values of the control output
value at different times are calculated, and a provisional value of
the control input value is calculated based on this evaluation
function value. This enables overshoot and oscillatory behavior of
the control output value relative to the target value to be further
suppressed for a control object having a large delay
characteristic.
[0038] According to the fifth aspect of the present invention, a
plant model containing a plurality of control input terms
proportional to a control input value and a plurality of control
output terms proportional to a control output value is established,
and the future predicted values of the control output value at a
plurality of different times are calculated by recursively using
output of this plant model in a control output term. In this way it
is possible to reduce the computational load by calculating future
predicted values of the control output value at a plurality of
different times with a recursive operation.
[0039] According to the sixth aspect of the invention, the future
predicted values of the control output value is calculated at a
plurality of different times by way of recursively using, in the
control output term, the output of the plant model using the
provisional value of the control input value not containing the
periodic reference signal in the control input term.
[0040] Herein, when recursively using the output of the plant model
as described above, the influence of the periodic reference signal
is accumulated, a result of which the convergence property of the
control output value to the target value may decline. According to
the present invention, the convergence property of the control
output value to the target value can be prevented from declining by
recursively using the output of the plant model using a provisional
value of the control input value not containing the reference
signal. Therefore, overshoot and oscillatory behavior of the
control output value relative to the target value can be further
suppressed.
[0041] According to the seventh aspect of the present invention,
the provisional value of the control input value containing the
periodic reference signal is only used in a portion of the
plurality of control input terms of the plant model.
[0042] Herein, in a case of the plant model having a plurality of
control input terms, an old control input value may be used in this
plurality of control input terms when calculating a relatively
close future predicted value. In this situation, if the provisional
value of the control input value containing the reference signal is
used in a plurality of control input terms, the influence of this
control input value containing the reference signal will be
redundant, a result of which the convergence property of the
control output value to the target value may decline. According to
the present invention, such a redundancy can be avoided and the
convergence property of the control output value to the target
value can be prevented from declining by only using the provisional
value of the control input value containing the reference signal in
a portion of the plurality of control input terms. Therefore,
overshoot and oscillatory behavior of the control output value
relative to the target value can be further suppressed.
[0043] According to the eighth aspect of the present invention, the
supply amount of the reducing agent can be controlled by the
controller so that the output value of the exhaust detection means
matches a predetermined target value. Such an exhaust purification
device using a selective reduction catalyst is a system having a
large delay in the response to the supply of reducing agent. With
the controller, it is possible to causes the output value of the
exhaust detection means to quickly and highly precisely match a
target value set so that the exhaust purification efficiency of the
exhaust purification device is optimal, even for such a system in
which the response delay is large. In this case as well, overshoot
and oscillatory behavior of the output value relative to the target
value can be suppressed. This enables the exhaust purification
performance of the exhaust purification device to be improved.
[0044] According to the ninth aspect of the present invention, the
target value of torque of the drive source can be controlled by the
controller so that the detected value of the drive wheel speed
detection means matches a predetermined target value. A vehicle
transfers torque generated by the drive source to the wheels
through devices having large inertial mass such as the driveshaft,
propeller shaft, crank, transmission. As a result, such a vehicle
is a system having a large response delay in relation to the
generated torque of the drive source. The revolution speed of a
drive wheel can be quickly and highly precisely caused to match a
predetermined target value by the controller, even for such a
system in which the response delay is large. In this case as well,
overshoot and oscillatory behavior of the detected value relative
to the target value can be suppressed. This enables the
acceleration performance and stability of the vehicle to be
improved.
BRIEF DESCRIPTION OF THE DRAWINGS
[0045] FIG. 1 is a schematic diagram showing configurations of an
ECU and an exhaust purification device of an engine controlled by
this ECU according to a first embodiment of the present
invention;
[0046] FIG. 2 presents graphs for illustrating the principle
deciding an FB injection amount according to the embodiment;
[0047] FIG. 3 presents graphs showing a relationship between a
detected ammonia concentration and a urea injection amount in a
case of performing urea injection control by way of conventional
feedback control;
[0048] FIG. 4 is a block diagram showing a configuration of a
module calculating the urea injection amount according to the
embodiment;
[0049] FIG. 5 is a block diagram showing a configuration of a
feedback controller according to the embodiment;
[0050] FIG. 6 presents graphs showing a relationship between an
evaluation function value and an urea injection amount difference
according to the embodiment;
[0051] FIG. 7 presents graphs showing a correlation between the
evaluation function value and a moving average value relative to
the optimum injection amount according to the embodiment;
[0052] FIG. 8 is a block diagram showing a configuration between a
conventional extremum searching optimization unit and a control
object;
[0053] FIG. 9 is a graph showing an example of a control map for
determining a FF injection amount according to the embodiment;
[0054] FIG. 10 is a flowchart showing a sequence of urea injection
control processing executed by the ECU according to the
embodiment;
[0055] FIG. 11 presents graphs showing relationships between the
detected ammonia concentration, the evaluation function value, and
the urea injection amount in a case of performing urea injection
control by way of the ECU according to the embodiment;
[0056] FIG. 12 is a schematic diagram showing configurations of an
ECU and a vehicle controlled by this ECU according to a second
embodiment of the present invention;
[0057] FIG. 13 is a graph showing a relationship between a torque
value and an engine revolution speed according to the
embodiment;
[0058] FIG. 14 is a graph showing a relationship between a
coefficient and an accelerator opening according to the
embodiment;
[0059] FIG. 15 presents graphs showing examples of control by the
ECU according to the embodiment;
[0060] FIG. 16 is a schematic diagram showing a configuration of a
conventional exhaust purification device;
[0061] FIG. 17 presents graphs showing relationships between a NOx
emission amount and ammonia emission amount downstream of a
selective reduction catalyst and the output of the NOx sensor in
the conventional exhaust purification device; and
[0062] FIG. 18 presents graphs showing relationships between the
NOx emission amount and ammonia emission amount downstream of the
selective reduction catalyst in a conventional exhaust purification
device using an ammonia sensor.
EXPLANATION OF REFERENCE NUMERALS
[0063] 1 engine (internal combustion engine) [0064] 2 exhaust
purification device (plant) [0065] 231 first selective reduction
catalyst (selective reduction catalyst) [0066] 25 urea injection
device (reducing agent supply means) [0067] 26 ammonia sensor
(exhaust detection means) [0068] 3 ECU (controller) [0069] 51
feedback controller [0070] 511 predictor (predicted value
calculating means) [0071] 512 evaluation function value calculating
unit (evaluation function value calculating means) [0072] 513
extremum searching optimization unit (extremum searching means)
[0073] 52 target value correction portion (target value correction
means) [0074] 54 adder (control input value calculating means)
[0075] 6 vehicle (plant) [0076] 61 drive wheel (wheel) [0077] 7 ECU
(controller) [0078] 71 feedback controller [0079] 712 predictor
(predicted value calculating means) [0080] 713 evaluation function
value calculating unit (evaluation function value calculating
means) [0081] 714 extremum searching optimization unit (extremum
searching means) [0082] 72 feedforward controller [0083] 74 target
value correction portion (target value correction means) [0084] 76
adder (control input value calculating means)
PREFERRED MODE FOR CARRYING OUT THE INVENTION
First Embodiment
[0085] Hereinafter, a first embodiment of the present invention
will be explained while referring to the drawings.
[0086] FIG. 1 is a schematic diagram showing configurations of an
electronic control unit (hereinafter referred to as "ECU") 3 as a
controller, and an exhaust purification device 2 of an internal
combustion engine (hereinafter referred to as "engine") 1 as a
plant controlled by this ECU 3, according to the first embodiment
of the present invention. The engine 1 is a gasoline engine of
lean-burn operating type or a diesel engine, and is mounted in a
vehicle, which is not illustrated.
[0087] The exhaust purification device 2 is configured to include
an oxidation catalyst 21 provided in an exhaust channel 11 of the
engine 1, a urea selective reduction catalyst 23 that is provided
in the exhaust channel 11 and purifies nitrogen oxides (hereinafter
referred to as "NOx") in exhaust flowing through this exhaust
channel 11 under the presence of a reducing agent, and a urea
injection device 25 that supplies urea water as a reducing agent
into the exhaust channel 11 on an upstream side of the urea
selective reduction catalyst 23, and is controlled by the ECU
3.
[0088] The urea injection device 25 includes a urea tank 251 and a
urea injection value 253.
[0089] The urea tank 251 stores urea water, and is connected to the
urea injection valve 253 via a urea supply pipe 254 and a urea
pump, which is not illustrated. A urea level sensor 255 is provided
to this urea tank 251. This urea level sensor 255 detects the water
level of the urea water in the urea tank 251, and outputs a
detection signal substantially proportional to this water level to
the ECU 3.
[0090] The urea injection valve 253 is connected to the ECU 3,
operates according to a control signal from the ECU 3, and injects
urea water into the exhaust channel 11 in accordance with this
control signal. In other words, urea injection control is
executed.
[0091] The oxidation catalyst 21 is provided more on an upstream
side in the exhaust channel 11 than the urea selective reduction
catalyst 23 and the urea injection valve 253, and converts NO in
the exhaust to NO.sub.2, thereby promoting the reduction of NOx in
the urea selective reduction catalyst 23.
[0092] The urea selective reduction catalyst 23 is configured to
include a first selective reduction catalyst 231, and a second
selective reduction catalyst 232 that is provided in the exhaust
channel 11 more on a downstream side than the first selective
reduction catalyst 231. This first selective reduction catalyst 231
and second selective reduction catalyst 232 each selectively reduce
NOx in exhaust under an atmosphere in which ammonia water is
present. More specifically, when urea water is injected by the urea
injection device 25, ammonia is produced from the urea in this
first selective reduction catalyst 231 and second selective
reduction catalyst 232, and the NOx in the exhaust is selectively
reduced by this ammonia.
[0093] In addition to an ammonia sensor 26 as an exhaust detection
means, catalyst temperature sensor 27 and NOx sensor 28, a crank
angle position sensor 14, accelerator opening sensor 15, and urea
remaining amount warning light 16 are connected to the ECU 3.
[0094] The ammonia sensor 26 detects the concentration of ammonia
(hereinafter referred to as "ammonia concentration") NH3.sub.CONS
of exhaust in the exhaust channel 11 between the first selective
reduction catalyst 231 and the second selective reduction catalyst
232, and supplies a detection signal substantially proportional to
the ammonia concentration NH3.sub.CONS thus detected to the ECU
3.
[0095] The catalyst temperature sensor 27 detects a temperature
(hereinafter referred to as "catalyst temperature") T.sub.SCR of
the first selective reduction catalyst 231, and supplies a
detection signal substantially proportional to the catalyst
temperature T.sub.SCR thus detected to the ECU 3.
[0096] The NOx sensor 28 detects a concentration of NOx in the
exhaust (hereinafter referred to as "NOx concentration")
NOX.sub.CONS flowing into the first selective reduction catalyst
231, and supplies a detection signal substantially proportional to
the NOx concentration NOX.sub.CONS thus detected to the ECU 3. More
specifically, the NOx sensor 28 detects the NOx concentration of
exhaust in the exhaust channel 11 on a more upstream side than a
position at which urea water is supplied from the urea injection
value 253.
[0097] The crank angle position sensor 14 detects a rotation angle
of the crank shaft of the engine 1 along with generating a pulse at
every 1.degree. of crank angle, and supplies this pulse signal to
the ECU 3. A revolution speed NE of the engine 1 is calculated by
the ECU 3 based on this pulse signal. The crank angle position
sensor 14 further generates a cylinder discriminating pulse at a
predetermined crank angle position of a specific cylinder, and
supplies this to the ECU 3.
[0098] The accelerator opening sensor 15 detects a depression
amount (hereinafter referred to as "accelerator opening") AP of the
accelerator pedal, which is not illustrated, of the vehicle, and
supplies a detection signal substantially proportional to the
accelerator opening AP thus detected to the ECU 3.
[0099] The urea remaining amount warning light 16 is provided in
the instrument panel of the vehicle, for example, and illuminates
in response to the remaining amount of urea water in the urea tank
251 having decreased past a predetermined remaining amount. With
this, the fact that the remaining amount of urea water in the urea
tank 251 has decreased is warned to the operator.
[0100] The ECU 3 is provided with an input circuit having functions
such as of shaping input signal waveforms from every kind of
sensor, correcting the voltage levels to predetermined levels, and
converting analog signal values to digital signal values, and a
central processing unit (hereinafter referred to as "CPU"). In
addition to this, the ECU 3 is provided with a storage circuit that
stores every kind of calculation program executed by the CPU,
calculation results, and the like, and an output circuit that
outputs control signals to the engine 1, urea injection valve 253,
and the like, and executes the urea injection control described
later.
[0101] Incidentally, the first selective reduction catalyst 231 and
the second selective reduction catalyst 232 of the aforementioned
urea selective reduction catalyst 23 each have a function of
reducing NOx in the exhaust with the ammonia produced from urea as
well as having a function of storing the ammonia produced in a
predetermined amount.
[0102] Hereinafter, the ammonia amount stored in the first
selective reduction catalyst 231 is defined as a first storage
amount, and the ammonia amount that can be stored in the first
selective reduction catalyst 231 is defined as a first storage
capacity. In addition, the ammonia amount stored in the second
selective reduction catalyst 232 is defined as a second storage
amount, and the ammonia amount that can be stored in the second
selective reduction catalyst 232 is defined as a second storage
capacity.
[0103] The ammonia stored in this way is consumed as appropriate in
the reduction of NOx in the exhaust. As a result, the NOx reduction
rate of the selective reduction catalysts 231 and 232 increases in
accordance with the first and second storage amounts increasing. In
addition, in a case such as when the supply amount of urea water is
small relative to the amount of NOx generated, the ammonia thus
stored is consumed in the reduction of NOx so as to compensate for
this deficient amount of urea water.
[0104] Herein, in a case of the ammonia produced exceeding the
storage capacity of the selective reduction catalyst 231 and 232,
respectively, the ammonia produced is discharged to downstream of
each of the selective reduction catalysts 231 and 232. The ammonia
not being stored in each of the selective reduction catalysts 231
and 232 and being discharged to downstream of each in this way is
referred to hereinafter as "ammonia slip".
[0105] In order to continuously maintain the NOx reduction rate of
such selective reduction catalysts 231 and 232 to be high, it is
preferable to continuously maintain a state in which ammonia is
stored in these selective reduction catalysts 231 and 231 in an
amount close to the storage capacity of each. However, in a state
in which ammonia of an amount near the storage capacity is stored
in this way, it is easy for ammonia slip to occur, and thus ammonia
may be discharged to outside the vehicle. In particular, it is
preferable for ammonia slip of the second selective reduction
catalyst 232 to be prevented as much as possible.
[0106] In light of the above such point, the ECU 3 of the present
embodiment executes urea injection control of the exhaust
purification device 3 based on a principle such as that shown
below. More specifically, as shown in the following formula (1), a
urea injection amount G.sub.UREA is shown as the sum of a feedback
injection amount (hereinafter referred to as "FB injection amount")
G.sub.UREA.sub.--.sub.FB and a predetermined reference injection
amount G.sub.UREA.sub.--.sub.BS.
G.sub.UREA=G.sub.UREA.sub.--.sub.FB+G.sub.UREA.sub.--.sub.BS
(1)
[0107] The FB injection amount G.sub.UREA.sub.--.sub.FB is a
control input value for controlling so that the output value of the
ammonia sensor 26, i.e. the detected ammonia concentration
NH3.sub.CONS, matches a predetermined target ammonia concentration
NH3.sub.CONS.sub.--.sub.TRGT and is determined according to the
detected ammonia concentration NH3.sub.CONS.
[0108] FIG. 2 presents graphs for illustrating the principle
determining the FB injection amount G.sub.UREA.sub.--.sub.FB. In
FIG. 2, the horizontal axis indicates time, and the vertical axis
indicates the NOx concentration downstream of the first selective
reduction catalyst 231, the ammonia concentration downstream of the
first selective reduction catalyst 231, and the detected ammonia
concentration NH3.sub.CONS, in order from the top.
[0109] The NOx reduction rate of the first selective reduction
catalyst 231 rises when increasing the urea injection amount
G.sub.UREA, since the ammonia produced in the first selective
reduction catalyst 231 also increases. As a result, the NOx
concentration downstream of the first selective reduction catalyst
231 decreases accompanying an increase in the urea injection amount
G.sub.UREA, as shown in the top graph of FIG. 2. In addition, when
the urea injection amount G.sub.UREA exceeds the injection amount
shown by the star, the NOx reduction rate of the first selective
reduction catalyst 231 becomes a maximum, and the NOx concentration
becomes substantially constant irrespective of the urea injection
amount G.sub.UREA. In other words, the urea water of an amount
exceeding the star is surplus relative to that reducing NOx in the
first selective reduction catalyst 231, and indicates being
discharged to the downstream side thereof. In addition,
accompanying this, the ammonia concentration of the exhaust
downstream of the first selective reduction catalyst 231 increases
when the urea injection amount G.sub.UREA exceeds the injection
amount shown by the star, as shown in the middle graph of FIG.
2.
[0110] More specifically, as shown in the bottom graph of FIG. 2,
the target ammonia concentration NH3.sub.CONS.sub.--.sub.TRGT is
set to a value somewhat larger than "0", and a target ammonia slip
range RNH3.sub.CONS.sub.--.sub.TRGT is defined by a lower limit
NH3.sub.CONS.sub.--.sub.LMTL that is less than this target ammonia
concentration NH3.sub.CONS.sub.--.sub.TRGT and an upper limit
NH3.sub.CONS.sub.--.sub.LMTH that is larger than this. Then, the
urea injection amount is controlled so that the output value of the
ammonia sensor 26, i.e. the detected ammonia concentration
NH3.sub.CONS, matches the target ammonia concentration
NH3.sub.CONS.sub.--.sub.TRGT within the range of this target
ammonia slip range RNH3.sub.CONS.sub.--.sub.TRGT. Herein, the
target ammonia slip range RNH3.sub.CONS.sub.--.sub.TRGT is
preferably set taking into account the detection resolution of the
ammonia sensor 26.
[0111] A state in which ammonia is saturated from the first
selective reduction catalyst 231 can be maintained by performing
control of the urea injection amount G.sub.UREA so that the
detected ammonia concentration NH3.sub.CONS matches the target
ammonia concentration NH3.sub.CONS.sub.--.sub.TRGT that is somewhat
larger than "0". With this, a high NOx reduction rate can be
maintained in the first selective reduction catalyst 231.
[0112] In addition, even if the reduction of NOx in the first
selective reduction catalyst 231 is insufficient, the NOx reduction
rate as an overall of the first selective reduction catalyst 231
and the second selective reduction catalyst 232 can be maintained
to be high by causing this remaining NOx and ammonia slipped to the
second selective reduction catalyst 232 to react. In addition,
ammonia slip by the second selective reduction catalyst 232 can
also be prevented from occurring by setting the target ammonia
concentration NH3.sub.CONS.sub.--.sub.TRGT to a value somewhat
larger than "0".
Problems of Urea Injection Control
[0113] Next, the problems that may occur in a case of performing
urea injection control by way of convention feedback control based
on the aforementioned such principle and problems observed by the
inventors of the present application will be explained in
detail.
[0114] FIG. 3 presents graphs showing a relationship between a
detected ammonia concentration NH3.sub.CONS and a urea injection
amount G.sub.UREA in a case of performing urea injection control by
way of conventional feedback control. In addition, in this example
shown in FIG. 3, after causing the target ammonia concentration
NH3.sub.CONS.sub.--.sub.TRGT to increase step-wise from "0" to a
predetermined value at time t.sub.0, modeling error and noise were
imparted at time t.sub.A, time t.sub.B and time t.sub.C.
[0115] At time t.sub.A, the volume of exhaust was made to change as
modeling error, with the NOx amount in the exhaust kept
constant.
[0116] At time t.sub.B, the NOx amount in the exhaust increased as
noise. It should be noted that main causes for the NOx amount in
the exhaust increasing may be the fuel injection timing being made
to be advanced than a reference, the injection pattern being
changed so that the diffusion combustion proportion increases, or
the like.
[0117] At time t.sub.C, the NOx amount in the exhaust decreases as
noise. It should be noted that the primary causes for the NOx
amount in the exhaust decreasing may be the fuel injection timing
being made to be more retarded than a reference, the injection
pattern being changed so that the premix combustion ratio
increases, or the like.
[0118] As shown in FIG. 3, overshoot occurs in the detected ammonia
concentration NH3.sub.CONS immediately after the target ammonia
concentration NH3.sub.CONS.sub.--.sub.TRGT is made to change
step-wise at time t.sub.0. In addition, immediately after overshoot
occurs, the detected ammonia concentration NH3.sub.CONS decreases
more than necessary, and the detected ammonia concentration
NH3.sub.CONS shows oscillatory behavior as a result. In addition,
immediately after applying noise or modeling error at times t.sub.B
and t.sub.C, the detected ammonia concentration NH3.sub.CONS shows
overshoot and oscillatory behavior relative to the target ammonia
concentration NH3.sub.CONS.sub.--.sub.TRGT. This behavior is
considered to originate from the fact that there is a large delay
characteristic between the urea injection amount G.sub.UREA and the
detected ammonia concentration NH3.sub.CONS.
[0119] If such an oscillatory behavior occurs and the detected
ammonia concentration NH3.sub.CONS exceeds the target ammonia
concentration NH3.sub.CONS.sub.--.sub.TRGT, the ideal balance
between the ammonia amount consumed in the reduction of NOx in the
second selective reduction catalyst and the ammonia amount emitted
from the first selective reduction catalyst to the second selective
reduction catalyst will be broken, and the ammonia storage amount
of the second selective reduction catalyst will gradually increase.
As described above, with the exhaust purification device of the
present embodiment provided with the two selective reduction
catalysts of the first selective reduction catalyst and the second
selective reduction catalyst, even if oscillatory behavior occurs
in the detected ammonia concentration NH3.sub.CONS, ammonia slip of
the second selective reduction catalyst will not occur immediately.
However, if such oscillatory behavior continues over an extended
period of time, ammonia slip of the second selective reduction
catalyst will also occur, and an unpleasant odor may emanate from
the tailpipe of the vehicle.
[0120] In addition, as shown in FIG. 3, the oscillatory behavior of
the detected ammonia concentration NH3.sub.CONS originates from the
oscillatory behavior of the urea injection amount G.sub.UREA. In
other words, it is shown that, at the times when the detected
ammonia concentration NH3.sub.CONS falls below the target ammonia
concentration NH3.sub.CONS.sub.--.sub.TRGT, the urea injection
amount immediately prior thereto had been decreased more than
necessary. The selective reduction catalyst has a characteristic
whereby the reduction rate of NOx declines considerably if the urea
injection amount G.sub.UREA decreases more than necessary.
Therefore, when oscillatory behavior occurs in the detected ammonia
concentration NH3.sub.CONS, the NOx reduction rate of the first
selective reduction catalyst may decline considerably.
[0121] In a case of performing urea injection control by
conventional feedback control as described above, there are
concerns such as of the occurrence of ammonia slip and a decline in
the NOx reduction rate due to the response delay between the urea
injection amount G.sub.UREA and the detected ammonia concentration
NH3.sub.CONS. A detailed configuration of the controller according
to the invention of the present application, which has been made
taking into account the above such problems of conventional
feedback control, will be explained hereinafter.
[0122] FIG. 4 is a block diagram showing the configuration of a
module calculating the urea injection amount G.sub.UREA. This
module controls the exhaust purification device 2, setting the urea
injection amount G.sub.UREA from the urea injection device as a
control input value and setting the detected ammonia concentration
NH3.sub.CONS from the ammonia sensor as a control output value. The
functions of this module are realized by processing executed by the
ECU 3.
[0123] The module shown in FIG. 4 is configured to include a
feedback controller 51 that calculates the FB injection amount
G.sub.UREA.sub.--.sub.FB, a target value correction portion 52 that
corrects the target ammonia concentration
NH3.sub.CONS.sub.--.sub.TRGT, an identifier 53 that identifies a
plurality of model parameters used by the feedback controller 51,
and a plurality of adders 54, 56, 57 and 58.
[0124] The feedback controller 51 includes a target value
correction unit 522, a predictor 511, an evaluation function value
calculating unit 512, and an extremum searching optimization unit
513, and calculates the FB injection amount
G.sub.UREA.sub.--.sub.FB so that the detected ammonia concentration
NH3.sub.CONS matches the target ammonia concentration
NH3.sub.CONS.sub.--.sub.TRGT.
[0125] For convenience of operation, this feedback controller 51
handles a detected ammonia concentration difference DNH3.sub.CONS
and a target ammonia concentration difference
DNH3.sub.CONS.sub.--.sub.TRGT obtained by subtracting the detected
ammonia concentration NH3.sub.CONS and the target ammonia
concentration NH3.sub.CONS.sub.--.sub.TRGT from a predetermined
reference value NH3.sub.CONS.sub.--.sub.BS, without directly
handling this detected ammonia concentration NH3.sub.CONS and
target ammonia concentration NH3.sub.CONS.sub.--.sub.TRGT.
[0126] This detected ammonia concentration difference DNH3.sub.CONS
and target ammonia concentration difference
DNH3.sub.CONS.sub.--.sub.TRGT are respectively calculated by the
adders 56 and 57, as shown in the following formulas (2) and
(3).
DNH3.sub.CONS=NH3.sub.CONS-NH3.sub.CONS.sub.--.sub.BS (2)
DNH3.sub.CONS.sub.--.sub.TRGT=NH3.sub.CONS.sub.--.sub.TRGT-NH3.sub.CONS.-
sub.--.sub.BS (3)
[0127] Herein, the reference value NH3.sub.CONS.sub.--.sub.BS in
the above formulas (2) and (3) is set to an average value or median
value of the target ammonia concentration
NH3.sub.CONS.sub.--.sub.TRGT or an arbitrary value within the range
of the target ammonia concentration NH3.sub.CONS.sub.--.sub.TRGT,
for example.
[0128] In other words, this feedback controller 51 calculates the
FB injection amount G.sub.UREA.sub.--.sub.FB so that the detected
ammonia concentration difference DNH3.sub.CONS matches the target
ammonia concentration difference DNH3.sub.CONS.sub.--.sub.TRGT.
[0129] The adder 54 calculates the urea injection amount G.sub.UREA
by way of adding the FB injection amount G.sub.UREA.sub.--.sub.FB
thus calculated and the reference injection amount
G.sub.UREA.sub.--.sub.BS as shown in the above formula (1), and
inputs this to the exhaust purification device 2 as a control input
value.
[0130] The adder 58 calculates a urea injection amount difference
DG.sub.UREA by subtracting a reference injection amount
G.sub.UREA.sub.--.sub.BS from the urea injection amount G.sub.UREA
thus calculated by the adder 54, a shown in the following formula
(4), and inputs this to the predictor 511 of the feedback
controller 51 and to the identifier 53.
DG.sub.UREA=G.sub.UREA-G.sub.UREA.sub.--.sub.BS (4)
[0131] In the present embodiment, although a feedforward injection
amount (hereinafter referred to as "FF injection amount")
G.sub.UREA.sub.--.sub.FF is used as this reference injection amount
G.sub.UREA.sub.--.sub.BS it is not limited thereto. The reference
injection amount G.sub.UREA.sub.--.sub.BS may be set to an
arbitrary constant including "0", or a urea injection amount for
maintaining the detected ammonia concentration NH3.sub.CONS to the
aforementioned reference value NH3.sub.CONS.sub.--.sub.BS.
Alternatively, the reference injection amount
G.sub.UREA.sub.--.sub.BS may be set to a learned value for
maintaining the detected ammonia concentration NH3.sub.CONS to
within a target ammonia slip range
RNH3.sub.CONS.sub.--.sub.TRGT.
[0132] As explained in detail later, the target value correction
portion 52 calculates a corrected target ammonia concentration
DNH3.sub.CONS.sub.--.sub.TRGT.sub.--.sub.MOD by conducting a
predetermined operation on the target ammonia concentration
difference DNH3.sub.CONS.sub.--.sub.TRGT in order to prevent
overshoot and oscillatory behavior of the detected ammonia
concentration difference DNH3.sub.CONS relative to the target
ammonia concentration DNH3.sub.CONS.sub.--.sub.TRGT.
[0133] As explained in detail later, the identifier 53 identifies a
plurality of model parameters a1, a2, b1 and b2 used by the
predictor 511 of the feedback controller 51.
[0134] It should be noted that an operation is performed by the
respective modules shown in FIG. 4 in different control cycles. In
the present embodiment, the control cycles of the adders 54, 56, 57
and 58, i.e. the update period of the urea injection amount
G.sub.UREA, urea injection amount difference DG.sub.UREA, detected
ammonia concentration NH3.sub.CONS, detected ammonia concentration
difference DNH3.sub.CONS, target ammonia concentration
NH3.sub.CONS.sub.--.sub.TRGT and target ammonia concentration
difference DNH3.sub.CONS.sub.--.sub.TRGT is defined as .DELTA.Ti.
In addition, the control cycle of the identifier 53 is also defined
as .DELTA.Ti. Therefore, the update period of the model parameters
a1, a2, b1 and b2 are also .DELTA.Ti. It should be noted that, in
the present embodiment, this update period .DELTA.Ti is set on the
order of 0.2 to 1.0 seconds, for example.
[0135] Hereinafter, the configuration of each module will be
explained in detail.
Configuration of Target Value Correction Portion
[0136] The configuration of the target value correction portion 52
will be explained.
[0137] The target value correction portion 52 is configured to
include a target value filter 521 and a target value correction
unit 522, and calculates a corrected target ammonia concentration
DNH3.sub.CONS.sub.--.sub.TRGT.sub.--.sub.MOD by the sequence shown
in the following formulas (5) to (8).
[0138] The target value filter 521 is a so-called first-order time
lag filter. In other words, the target value filter 521 calculates
a virtual target value DNH3.sub.CONS.sub.--.sub.TRGT.sub.--.sub.F
by performing the operations shown in the following formulas (5)
and (6) with respect to the target ammonia concentration difference
DNH3.sub.CONS.sub.--.sub.TRGT(k) thus calculated.
DNH3.sub.CONS.sub.--.sub.TRGT(k+i)=DNH3.sub.CONS.sub.--.sub.TRGT(k)
(i=0.about.np) (5)
DNH3.sub.CONS.sub.--.sub.TRGT.sub.--.sub.F(k+i)=-POLE.sub.FDN3.sub.CONS.-
sub.--.sub.TRGT.sub.--.sub.F(k+i-1)+(1+POLE.sub.F)DNH3.sub.CONS.sub.--.sub-
.TRGT(k+i) (i=0.about.np) (6)
[0139] Herein, the notation "k" is a notation indicating
discretized time and, in the present embodiment, indicates being
data detected or calculated every model sampling period .DELTA.Tm
described later. In other words, in a case of the notation "k"
being data detected or calculated in a current control time, the
notation "k-1", for example, indicates being data detected or
calculated in a previous control time, i.e. at .DELTA.Tm prior. It
should be noted that the notation "k" is omitted as appropriate in
the following explanation.
[0140] In addition, in the above formulas (5) and (6), POLE.sub.F
is a tracking characteristic designating parameter for designating
the tracking characteristic of the detected ammonia concentration
difference DNH3.sub.CONS relative to the target ammonia
concentration difference DNHT3.sub.CONS.sub.--.sub.TRGT and
POLE.sub.F set to be greater than -1 and less than 0. In addition,
in the above formulas (5) and (6), np is a positive integer and
indicates a prediction step number. The prediction step number np
indicates the step number being predicted (step number for which
.DELTA.Tm is set as the control cycle) in the predictor 511, as
explained later while referring to formula (17).
[0141] The target value correction unit 522 calculates the
corrected target ammonia concentration
DNH3.sub.CONS.sub.--.sub.TRGT.sub.--.sub.MOD according to the
response delay characteristic of the detected ammonia concentration
difference DNH3.sub.CONS as shown in the following formulas (7) and
(8), in order to prevent overshoot and oscillatory behavior of the
detected ammonia concentration DNH3.sub.CONS relative to the target
ammonia concentration DNH3.sub.CONS.sub.--.sub.TRGT.
[0142] More specifically, this target value correction unit 522
first calculates deviation Ec between the detected ammonia
concentration difference DNH3.sub.CONS and the virtual target value
DNH3.sub.CONS.sub.--.sub.TRGT.sub.--.sub.F of the above formula
(6), as shown in the following formula (7).
Ec(k)=DNH3.sub.CONS(k)-DNH3.sub.CONS.sub.--.sub.TRGT.sub.--.sub.F(k)
(7)
[0143] Next, as shown in the following formula (8), the corrected
target ammonia concentration
DNH3.sub.CONS.sub.--.sub.TRGT.sub.--.sub.MOD is calculated by
adding a damping term that is proportional to the deviation Ec to
the virtual target value
DNH3.sub.CONS.sub.--.sub.TRGT.sub.--.sub.F.
DNH3.sub.CONS.sub.--.sub.TRGT.sub.--.sub.MOD(k+i)=DNH3.sub.CONS.sub.--.s-
ub.TRGT.sub.--.sub.F(k)+(-POLE.sub.E).sup.iEc(k) (8)
[0144] Herein, POLE.sub.E, which is contained in the damping term
of the above formula (8), is a deviation convergence characteristic
designating parameter for designating the convergence
characteristic of the deviation Ec, and POLE.sub.E is set to be
greater than -1 and less than 0.
[0145] In the above way, the corrected target ammonia concentration
DNH3.sub.CONS.sub.--.sub.TRGT.sub.--.sub.MOD is set between the
target ammonia concentration difference
DNH3.sub.CONS.sub.--.sub.TRGT and the detected ammonia
concentration difference DNH3.sub.CONS, taking into account of the
response delay characteristic of the detected ammonia concentration
difference DNH3.sub.CONS.
[0146] Herein, the corrected target ammonia concentration
DNH3.sub.CONS.sub.--.sub.TRGT.sub.--.sub.MOD indicates a target
value of a future time "k+1" relative to the present time "k", as
shown in the above formula (8). Therefore, as explained later while
referring to FIG. 6, this corrected target ammonia concentration
DNH3.sub.CONS.sub.--.sub.TRGT.sub.--.sub.MOD (k+1) is more strictly
set between a target ammonia concentration difference
DNH3.sub.CONS.sub.--.sub.TRGT (k+1) at a future time and a future
predicted value of the detected ammonia concentration difference
DNH3.sub.CONS (predicted ammonia concentration PREDNH3.sub.EXS
(k+1) described later).
Configuration of Identifier
[0147] The configuration of the identifier 53 will be
explained.
[0148] The identifier 53 identifies, by way of the recursive
identification algorithms shown in the following formulas (9) to
(15), the plurality of model parameters a1, a2, b1 and b2 of a
plant model used by the predictor 511 described later. More
specifically, a recursive identification algorithm is an algorithm
that calculates a current control value .theta.(m) of a model
parameter vector based on a latest control value .zeta.(m) of
processing object data and a previous control value .theta.(m-1) of
the model parameter vector.
[0149] .theta.(m) and .zeta.(m) are defined as four-component
vectors as shown in the following formulas (9) and (10).
.theta..sup.T(m)=[a1(m)a2(m)b1(m)b2(m)] (9)
.xi..sup.T(m)=[DNH3.sub.CONS(m-Mmi)DNH3.sub.CONS(m-2Mmi)DG.sub.UREA(m-Mm-
i)DG.sub.UREA(m-2Mmi)] (10)
[0150] Herein, the notation "m" is a notation indicating
discretized time, and indicates being data detected or calculated
every aforementioned control cycle .DELTA.Ti. In addition, this
control cycle and model sampling period .DELTA.Tm satisfy the
following formula (11).
.DELTA.Tm=Mmi.DELTA.Ti (11)
[0151] According to the recursive identification algorithm, the
current control value .theta.(m) of the model parameter vector
defining the model parameters a1, a2, b1 and b2 as components is
calculated from the previous control value .theta.(m-1) by the
following formula (12).
.theta.(m)=.theta.(m-1)+KP(m)ide(m) (12)
[0152] Herein, ide(m) in the above formula (12) indicates the
identification error defined by the following formula (13). KP(m)
indicates a gain coefficient vector defined by the following
formula (14). P(m) in the following formula (14) is a fourth-order
square matrix defined by the following formula (15). In addition,
in the following formula (15), I indicates a fourth-order unit
matrix, and .lamda.1 and .lamda.2 respectively indicate weighting
parameters.
ide(m)=.theta..sup.T(m-1).zeta.(m) (13)
KP ( m ) = P ( m ) .zeta. ( m ) 1 + .zeta. ( m ) T P ( m ) .zeta. (
m ) ( 14 ) P ( m + 1 ) = 1 .lamda.1 { 1 - .lamda. 2 P ( m ) .zeta.
( m ) .zeta. ( m ) T .lamda.1 + .lamda.2.zeta. ( m ) T P ( m )
.zeta. ( m ) } P ( m ) ( 15 ) ##EQU00001##
[0153] Herein, various recursive identification algorithms can be
selected as shown below, by setting the weighting parameters
.lamda.1 and .lamda.2 of the above formula (15) as follows.
[0154] .lamda.1=1, .lamda.2=0: fixed gain algorithm
[0155] .lamda.1=1, .lamda.2=1: least-squares method algorithm
[0156] .lamda.1=1, .lamda.2=.lamda.: gradual decrease gain
algorithm (0<.lamda.<1)
[0157] .lamda.1=.lamda., .lamda.2=1: weighted least-squared method
algorithm (0<.lamda.<1)
[0158] In addition, when calculating the model parameters a1, a2,
b1 and b2 based on the above formulas (9) to (15), the values of
these model parameters may gradually shift from the desired values.
In this case, a constraint condition may be imposed on these model
parameters. A specific example thereof will be explained later
while referring to formulas (37) and (38).
[0159] By occasionally identifying the model parameters a1, a2, h1
and b2 by way of the identifier 53 in the above way, it is possible
to prevent the control precision from declining, even in a case of
modeling error arising in the plant model described later due to
individual variation or degradation over time of the exhaust
purification device.
Configuration of Feedback Controller
[0160] The configuration of the feedback controller 51 will be
explained.
[0161] FIG. 5 is a block diagram showing the configuration of the
feedback controller 51.
[0162] The feedback controller 51 is configured to include the
aforementioned target value correction unit 522 that calculates the
corrected target ammonia concentration
DNH3.sub.CONS.sub.--.sub.TRGT.sub.--.sub.MOD, the predictor 511
that calculates the predicted ammonia concentration PREDNH3.sub.EXS
as a future predicted value of the detected ammonia concentration
difference DNH3.sub.CONS, the evaluation function value calculating
unit 512 that calculates an evaluation function value J based on
this corrected target ammonia concentration
DNH3.sub.CONS.sub.--.sub.TRGT.sub.--.sub.MOD and predicted ammonia
concentration PREDNH3.sub.EXS, and the extremum searching
optimization unit 513 that calculates an optimum injection amount
DG.sub.UREA.sub.--.sub.OPT and search input
DG.sub.UREA.sub.--.sub.EXS such that the evaluation function value
J thus calculated becomes an extremum, i.e. a minimum value.
[0163] In other words, in a case of regarding the target value
correction unit 522, predictor 511 and evaluation function value
calculating unit 512 as a virtual plant 515 as shown in FIG. 5, the
extremum searching optimization unit 513 can be deemed a controller
that calculates the optimum injection amount
DG.sub.UREA.sub.--.sub.OPT and search input
DG.sub.UREA.sub.--.sub.EXS so that the control output of this
virtual plant 515, i.e. evaluation function value J, becomes a
minimum value.
[0164] Herein, the predictor 511 and target value correction
portion 52, which respectively calculate the predicted ammonia
concentration PREDNH3.sub.EXS and corrected target ammonia
concentration DNH3.sub.CONS.sub.--.sub.TRGT.sub.--.sub.MOD
necessary in order to calculate the evaluation function value J,
are also included in the virtual plant 515. In addition, the
predicted ammonia concentration PREDNH3.sub.EXS changes depending
on the model parameters a1, a2, b1 and b2, detected ammonia
concentration difference DNH3.sub.CONS and urea injection amount
difference DG.sub.UREA, and the corrected target ammonia
concentration DNH3.sub.CONS.sub.--.sub.TRGT.sub.--.sub.MOD also
changes depending on the detected ammonia concentration difference
DNH3.sub.CONS and virtual target value
DNH3.sub.CONS.sub.--.sub.TRGT.sub.--.sub.F. As a result, the
detected ammonia concentration difference DNH3.sub.CONS, virtual
target value DNH3.sub.CONS.sub.--.sub.TRGT.sub.--.sub.F, urea
injection amount difference DG.sub.UREA and model parameters a1,
a2, b1 and b2 can be deemed as schedule parameters of the virtual
plant 515.
[0165] In addition, the extremum searching optimization unit 513
searches for a minimum value of the evaluation function value J
using a periodic reference signal S.sub.REF as described in detail
later. Then, the extremum searching optimization unit 513 inputs
the optimum injection amount DG.sub.UREA.sub.--.sub.OPT not
containing the component of this reference signal S.sub.REF and a
search input DG.sub.UREA.sub.--.sub.EXS containing the component of
the reference signal S.sub.REF to the predictor 511 of the
aforementioned virtual plant 515. In addition, the optimum
injection amount DG.sub.UREA.sub.--.sub.OPT, which does not contain
the component of the reference signal S.sub.REF, is the FB
injection amount G.sub.UREA.sub.--.sub.FB.
[0166] Moreover, the control cycle of this feedback controller 51
is basically .DELTA.Te, which is shorter than the update period
.DELTA.Ti of the urea injection amount G.sub.UREA. Therefore, the
update period .DELTA.Te of this optimum injection amount
DG.sub.UREA.sub.--.sub.OPT and search input
DG.sub.UREA.sub.--.sub.EXS is shorter than the update period
.DELTA.Ti of the urea injection amount G.sub.UREA. The details of
this control cycle .DELTA.Te will be described in detail later.
[0167] Hereinafter, the configurations of this predictor 511,
evaluation function value calculating unit 512 and extremum
searching optimization unit 513 will be explained in order.
Configuration of Predictor
[0168] The configuration of the predictor 511 will be
explained.
[0169] The predictor 511 calculates the predicted ammonia
concentration PREDNH3.sub.EXS based on the ammonia concentration
difference DG.sub.UREA, optimum injection amount
DG.sub.UREA.sub.--.sub.OPT, search input DG.sub.UREA.sub.--.sub.EXS
containing the component of the reference signal S.sub.REF, and the
urea injection amount difference DG.sub.UREA. In addition, when
calculating the predicted ammonia concentration PREDNH3.sub.EXS
this predictor 511 uses a model of the exhaust purification device
(hereinafter referred to as "plant model") showing the dynamic
characteristic of the detected ammonia concentration difference
DNH3.sub.CONS from the urea injection amount difference
DG.sub.UREA.
[0170] Hereinafter, an optimum form of the plant model will be
constructed while considering the problems thereof. It should be
noted that this predictor 511 uses the plant model described later
shown in formulas (27) to (35) as a final form.
[0171] First, as a plant model showing the dynamic characteristic
of the detected ammonia concentration difference DNH3.sub.CONS from
the urea injection amount difference DG.sub.UREA, a model such as
that shown in the following formula (16) is considered that
contains a plurality of control output terms (terms proportional to
the model parameters a1 and a2) proportional to the detected
ammonia concentration difference DNH3.sub.CONS and a plurality of
control input terms (terms proportional to the model parameters b1
and b2) proportional to the detected ammonia concentration
difference DNH3.sub.CONS.
DNH 3 CONS ( k + 1 ) = a 1 DNH 3 CONS ( k ) + a 2 DNH 3 CONS ( k -
1 ) + b 1 DG UREA ( k ) + b 2 DG UREA ( k - 1 ) ( 16 )
##EQU00002##
[0172] As described above, the notation "k" indicates being data
detected or calculated every model sampling period .DELTA.Tm. In
particular, this model sampling period .DELTA.Tm is preferably set
to be longer than the aforementioned update period .DELTA.Ti of the
urea injection amount G.sub.UREA. The reason thereof will be
explained.
[0173] With this predictor 511, the predicted ammonia concentration
PREDNH3.sub.EXS is calculated from the current time until after a
predetermined predicted time Tp, by way of recursively using the
output of the plant model, as will be described in detail later.
Herein, in a case of setting the prediction time Tp to a few to
several tens of seconds, for example, if the model sampling period
.DELTA.Tm is made a comparable level to .DELTA.Ti (approximately
0.2 to 1.0 seconds), the number of times recursively using the
plant model will amount to on the order of several tens of times,
the prediction error will accumulate due to modeling error, and the
prediction accuracy will decline.
[0174] In addition, taking this point into account, the prediction
time Tp and the model sampling period .DELTA.Tm are set so as to
satisfy the following formula (17). In other words, the prediction
time Tp is set so as to be equivalent to a few steps to a several
tens of steps in a case of performing operations under the model
sampling period .DELTA.Tm. In addition, herein, the step number
corresponding to the prediction time Tp is defined as the
prediction step number np in a case of performing operations under
the model sampling period .DELTA.Tm, as shown in the following
formula (17). In this case, the number of prediction steps np is
set from 1 to tens of steps.
TP/.DELTA.Tm=np (17)
[0175] In addition, in a case of defining the dead time of the
plant model as d, it is unnecessary to consider the dead time of
the model, and it is possible to perform stable control by setting
the control cycle .DELTA.Tm to be larger than the dead time d.
Therefore, the model sampling period .DELTA.Tm is preferably set as
shown in the following formula (18).
.DELTA.Tm>d (18)
[0176] More specifically, a time constant contained in the delay
characteristic between the urea injection amount G.sub.UREA and the
detected ammonia concentration NH3.sub.CONS of the exhaust
purification device is several seconds to tens of seconds; whereas,
the dead time is on the order to 1 to 2 seconds. Therefore, the
model sampling period .DELTA.Tm is set to approximately 1.0 to 3.0
seconds, for example.
[0177] Next, the detected ammonia concentration differences
DNH3.sub.CONS(k+1), DNH3.sub.CONS(k+2), . . . , DNH3.sub.CONS(k+np)
until after the prediction step number np are calculated as shown
in the following formulas (20) to (24), by recursively using the
output of the plant model shown in the above formula (16), i.e.
left side of the above formula (16), in the control output term of
the plant model during a subsequent step, and these are defined as
predicted ammonia concentrations PREDNH3.sub.CONS(k+1),
PREDNH3.sub.CONS(k+2), . . . , PREDNH3.sub.CONS(k+np) at each
time.
[0178] Herein, the future values of the urea injection amount
difference DG.sub.UREA contained in the control input term of the
plant model during a subsequent step are all set to values equal to
the urea injection amount difference DG.sub.UREA (k), as shown in
the following formula (19).
DG.sub.UREA(k+1)=DG.sub.UREA(k+2)= . . .
=DG.sub.UREA(k+np-1)=DG.sub.UREA(k) (19)
[0179] Predicted ammonia concentration PREDNH3.sub.CONS(k+1) after
1--model sample time
PREDNH 3 CONS ( k + 1 ) = a 1 DNH 3 CONS ( k ) + a 2 DNH 3 CONS ( k
- 1 ) + b 1 DG UREA ( k ) + b 2 DG UREA ( k - 1 ) ( .apprxeq. DNH 3
CONS ( k + 1 ) ) ( 20 ) ##EQU00003##
[0180] Predicted ammonia concentration PREDNH3.sub.CONS(k+2) after
2--model sample times
PREDNH 3 CONS ( k + 2 ) = a 1 DNH 3 CONS ( k + 1 ) + a 2 DNH 3 CONS
( k ) + b 1 DG UREA ( k + 1 ) + b 2 DG UREA ( k ) = a 1 PREDNH 3
CONS ( k + 1 ) + a 2 DNH 3 CONS ( k ) + b 1 DG UREA ( k ) + b 2 DG
UREA ( k ) ( .apprxeq. DNH 3 CONS ( k + 2 ) ) ( 21 )
##EQU00004##
[0181] Predicted ammonia concentration PREDNH3.sub.CONS(k+3) after
3--model sample times
PREDNH 3 CONS ( k + 3 ) = a 1 DNH 3 CONS ( k + 2 ) + a 2 DNH 3 CONS
( k + 1 ) + b 1 DG UREA ( k + 2 ) + b 2 DG UREA ( k + 1 ) = a 1
PREDNH 3 CONS ( k + 2 ) + PREDNH 3 CONS ( k + 1 ) + b 1 DG UREA ( k
) + b 2 DG UREA ( k ) ( .apprxeq. DNH 3 CONS ( k + 3 ) ) ( 22 )
##EQU00005##
[0182] Predicted ammonia concentration PREDNH3.sub.CONS(k+4) after
4--model sample times
PREDNH 3 CONS ( k + 4 ) = a 1 DNH 3 CONS ( k + 3 ) + a 2 DNH 3 CONS
( k + 2 ) + b 1 DG UREA ( k + 3 ) + b 2 DG UREA ( k + 4 ) = a 1
PREDNH 3 CONS ( k + 3 ) + PREDNH 3 CONS ( k + 2 ) + b 1 DG UREA ( k
) + b 2 DG UREA ( k ) ( .apprxeq. DNH 3 CONS ( k + 4 ) ) ( 23 )
##EQU00006##
[0183] predicted ammonia concentration PREDNH3.sub.CONS(k+np) after
np--model sample times
PREDNH 3 CONS ( k + np ) = a 1 DNH 3 CONS ( k + np - 1 ) + a 2 DNH
3 CONS ( k + np - 1 ) + b 1 DG UREA ( k + np - 2 ) + b 2 DG UREA (
k + np - 2 ) = a 1 PREDNH 3 CONS ( k + np - 1 ) + PREDNH 3 CONS ( k
+ np - 2 ) + b 1 DG UREA ( k ) + b 2 DG UREA ( k ) ( .apprxeq. DNH
3 CONS ( k + np ) ) ( 24 ) ##EQU00007##
[0184] Next, the problems of the predicted ammonia concentrations
PREDNH3.sub.CONS shown in the above formulas (20) to (24) derived
in the above way will be considered.
[0185] In the aforementioned way, the extremum searching
optimization unit 513 searches for the extremum of the evaluation
function value J using the periodic reference signal S.sub.REF.
However, the component of the reference signal S.sub.REF is not
included in the above formulas (20) to (24). As a result, the
correlation between the evaluation function value J and the
reference signal S.sub.REF cannot be determined, since the
influence of the reference signal S.sub.REF is not reflected in the
evaluation function value J that is calculated based on such a
predicted ammonia concentration PREDNH3.sub.CONS. Therefore, the
above formulas (20) to (24) are preferably changed so that the
component of the reference signal S.sub.REF is included
therein.
[0186] In addition, the model sampling period .DELTA.Tm is
preferably set to be longer than the update period .DELTA.Ti of the
urea injection amount G.sub.UREA, as described above. However, when
searching for a minimum value of the evaluation function value J,
and calculating the optimum injection amount
DG.sub.UREA.sub.--.sub.OPT and search input
DG.sub.UREA.sub.--.sub.EXS by way of the extremum searching
optimization unit 513, this optimum injection amount
DG.sub.UREA.sub.--.sub.OPT and search input
DG.sub.UREA.sub.--.sub.EXS preferably sufficiently converge within
the update period .DELTA.Ti of the urea injection amount
G.sub.UREA. As a result, it is necessary to change the above
formulas (20) to (24) derived in the model sampling period
.DELTA.Tm to be under the update period .DELTA.Te, which is shorter
than the update period .DELTA.Ti of the urea injection amount
G.sub.UREA.
[0187] Herein, these update periods .DELTA.Ti and .DELTA.Te and the
model sampling period .DELTA.Tm are periods satisfying the
following formulas (25) and (26). More specifically, the update
period .DELTA.Te is set to approximately 0.1 to 0.5 seconds, for
example, in a case of the update period .DELTA.Ti being set to
approximately 0.2 to 1.0 seconds, as explained above.
.DELTA.Ti=Mie.DELTA.Te (25)
.DELTA.Tm=Mmi.DELTA.Ti=MmiMie.DELTA.Te (26)
[0188] Taking the above such two problems into account, the above
formulas (20) to (24) are changed as shown in the following
formulas (27) to (35).
[0189] Predicted ammonia concentration PREDNH3.sub.EXS (n+MmiMie)
after 1--model sample time
PREDNH 3 EXS ( n + MmiMie ) = a 1 ( n , k - 1 ) DNH 3 CONS ( n , k
) + a 2 ( n , k - 1 ) DNH 3 CONS ( n - MmiMie , k ) + b 1 ( n , k -
1 ) DG UREA_EXS ( n ) + b 2 ( n , k - 1 ) DG UREA ( n , k - 1 )
.apprxeq. PREDNH 3 CONS ( k + 1 ) ( 27 ) ##EQU00008##
[0190] Predicted ammonia concentration PREDNH3.sub.EXS (n+2MmiMie)
after 2--model sample times
PREDNH 3 CNT ( n + MmiMie ) = a 1 ( n , k - 1 ) DNH 3 CONS ( n , k
) + a 2 ( n , k - 1 ) DNH 3 CONS ( n - MmiMie , k ) + b 1 ( n , k -
1 ) DG UREA_OPT ( n ) + b 2 ( n , k - 1 ) DG UREA ( n , k - 1 )
.apprxeq. PREDNH 3 CONS ( k + 1 ) ( 28 ) PREDNH 3 EXS ( n + 2
MmiMie ) = a 1 ( n , k - 1 ) PREDNH 3 CONS ( n + MmiMie ) + a 2 ( n
, k - 1 ) DNH 3 CONS ( n , k ) + b 1 ( n , k - 1 ) DG UREA_EXS ( n
) + b 2 ( n , k - 1 ) DG UREA_OPT ( n ) .apprxeq. PREDNH 3 CONS ( k
+ 2 ) ( 29 ) ##EQU00009##
[0191] Predicted ammonia concentration PREDNH3.sub.EXS (n+3MmiMie)
after 3--model sample times
PREDNH 3 CNT ( n + 2 MmiMie ) = a 1 ( n , k - 1 ) PREDNH 3 CNT ( n
+ MmiMie ) + a 2 ( n , k - 1 ) DNH 3 CONS ( n , k ) + b 1 ( n , k -
1 ) DG UREA_OPT ( n ) + b 2 ( n , k - 1 ) DG UREA_OPT ( n )
.apprxeq. PREDNH 3 CONS ( k + 2 ) ( 30 ) PREDNH 3 EXS ( n + 3
MmiMie ) = a 1 ( n , k - 1 ) PREDNH 3 CNT ( n + 2 MmiMie ) + a 2 (
n , k - 1 ) PREDNH 3 CONS ( n + MmiMie ) + b 1 ( n , k - 1 ) DG
UREA_EXS ( n ) + b 2 ( n , k - 1 ) DG UREA_OPT ( n ) .apprxeq.
PREDNH 3 CONS ( k + 3 ) ( 31 ) ##EQU00010##
[0192] Predicted ammonia concentration PREDNH3.sub.EXS (n+4MmiMie)
after 4--model sample times
PREDNH 3 EXS ( n + 3 MmiMie ) = a 1 ( n , k - 1 ) PREDNH 3 CNT ( n
+ 2 MmiMie ) + a 2 ( n , k - 1 ) PREDNH 3 CNT ( n + MmiMie ) + b 1
( n , k - 1 ) DG UREA_OPT ( n ) + b 2 ( n , k - 1 ) DG UREA_OPT ( n
) .apprxeq. PREDNH 3 CONS ( k + 3 ) ( 32 ) PREDNH 3 EXS ( n + 4
MmiMie ) = a 1 ( n , k - 1 ) PREDNH 3 CNT ( n + 3 MmiMie ) + a 2 (
n , k - 1 ) PREDNH 3 CNT ( n + 2 MmiMie ) + b 1 ( n , k - 1 ) DG
UREA_EXS ( n ) + b 2 ( n , k - 1 ) DG UREA_OPT ( n ) .apprxeq.
PREDNH 3 CONS ( k + 4 ) ( 33 ) ##EQU00011##
[0193] Predicted ammonia concentration PREDNH3.sub.EXS (n+npMmiMie)
after np--model sample time
PREDNH 3 EXS ( n + ( np - 1 ) MmiMie ) = a 1 ( n , k - 1 ) PREDNH 3
CNT ( n + ( np - 2 ) MmiMie ) + a 2 ( n , k - 1 ) PREDNH 3 CNT ( n
+ ( np - 3 ) MmiMie ) + b 1 ( n , k - 1 ) DG UREA_OPT ( n ) + b 2 (
n , k - 1 ) DG UREA_OPT ( n ) .apprxeq. PREDNH 3 CONS ( k + np - 1
) ( 34 ) PREDNH 3 EXS ( n + np MmiMie ) = a 1 ( n , k - 1 ) PREDNH
3 CNT ( n + ( np - 1 ) MmiMie ) + a 2 ( n , k - 1 ) PREDNH 3 CNT (
n + ( np - 2 ) MmiMie ) + b 1 ( n , k - 1 ) DG UREA_EXS ( n ) + b 2
( n , k - 1 ) DG UREA_OPT ( n ) .apprxeq. PREDNH 3 CONS ( k + np )
( 35 ) ##EQU00012##
[0194] Herein, the notation "n" is a notation indicating
discretized time, and indicates being data detected or calculated
every aforementioned control cycle .DELTA.Te.
[0195] In the above formulas (27) to (35), the model parameters
a1(n,k), a2(n,k), b1(n,k) and b2(n,k), urea injection amount
difference DG.sub.UREA(m,k), and detected ammonia concentration
DNH3.sub.CONS(n, k) respectively indicate parameters updated at the
period .DELTA.Ti as explained above, and over sampled at the
shorter time period .DELTA.Te.
[0196] In addition, the value of a previous control time, i.e.
(k-1), is used in the model parameters and the urea injection
amount difference DG.sub.UREA, as shown in the above formula (27).
Since the update period of the model parameters and DG.sub.UREA is
.DELTA.Ti, if the urea injection amount difference DG.sub.UREA is
calculated after the operation of the identifier, the value of the
previous control time (k-1) will basically be used in these model
parameters and urea injection amount difference DG.sub.UREA.
Herein, if the operation of the identifier is performed immediately
after the operation of the urea injection amount difference, for
example, the value of the current control time (k) will be used in
the model parameters, and the value of the previous control time
(k-1) will be used in the urea injection amount difference
DG.sub.UREA.
[0197] Characteristics of the above formulas (27) to (35) will be
explained herein.
[0198] As shown in the above formulas (27), (29), (31), (33) and
(35), the search input DG.sub.UREA.sub.--.sub.EXS containing the
reference signal is used only in a portion of the plurality of
control input terms (terms proportional to b1 and b2). More
specifically, among the two control input terms proportional to b1
and b2, the search input DG.sub.UREA.sub.--.sub.EXS is only used in
the term proportional to b1.
[0199] As shown in the above formulas (27) to (35), in the
predicted ammonia concentration PREDNH3.sub.EXS of a certain time,
the predicted ammonia concentration PREDNH3.sub.EXS one prior to
the model sample time of this time is used. As a result, if the
search input DG.sub.UREA.sub.--.sub.EXS containing the reference
signal were used in the term proportional to b2, the influence of
the search input DG.sub.UREA.sub.--.sub.EXS containing the
reference signal would be redundant between the term proportional
to b1 and the term proportional to b2, and it would become
difficult to determine the correlation between the reference signal
and the evaluation function value. Therefore, the convergence
property of the detected ammonia concentration NH3.sub.CONS to the
target ammonia concentration NH3.sub.CONS.sub.--.sub.TRGT may
decline as a result. In the present embodiment, the search input
DG.sub.UREA.sub.--.sub.EXS is only used in the term proportional to
b1 as explained above in order to prevent such a decline in the
convergence property.
[0200] In addition, as shown in the above formulas (27), (29),
(31), (33) and (35), among the predicted ammonia concentrations
PREDNH3.sub.EXS of certain times, a predicted ammonia concentration
PREDNH3.sub.CNT of one model sample time prior is recursively used
in the control output terms (terms proportional to a1 and a2).
Herein, as shown in the above formulas (28), (30), (32) and (34),
for these predicted ammonia concentrations PREDNH3.sub.CNT of one
model sample time prior recursively substituted, the optimum
injection amount DG.sub.UREA.sub.--.sub.OPT not containing the
reference signal is used in the control input terms (terms
proportional to b1 and b2).
[0201] Herein, if the search input DG.sub.UREA.sub.--.sub.EXS were
used in such predicted ammonia concentrations PREDNH3.sub.CNT of
one model sample time prior, the influence of the reference signal
would accumulate as the prediction time progresses and the control
system would easily become oscillatory, a result of which the
convergence property of the detected ammonia concentration
NH3.sub.CONS relative to the target ammonia concentration
NH3.sub.CONS.sub.--.sub.TRGT may decline.
[0202] In addition, even if such oscillation did not occur, if the
prediction time were changed, for example, the severity of the
reference signal on the evaluation function value J may change
greatly. In such a case, the gain of the controller also must be
reset, and a problem arises in that setting of the parameters
becomes complex. In order to avoid the above such problem in the
present embodiment, only the optimum injection amount
DG.sub.UREA.sub.--.sub.OPT not containing the reference signal is
used in such predicted ammonia concentrations PREDNH3.sub.CNT of
one model sample time prior recursively substituted.
[0203] Furthermore, as explained above, if the dynamic
characteristic of the predicted ammonia concentration of each model
sample time is unstable due to the predicted ammonia concentration
of each model sample time being recursively used in the control
output terms (terms proportional to a1 and a2) of a subsequent
model sample time in the above formulas (27) to (35), the predicted
ammonia concentration may diverge. Consequently, when identifying
the model parameters a1 and a2 by way of the identifier 53, these
model parameters a1 and a2 are identified under constraint
conditions such as those shown in the following formulas (37) and
(38).
|a1(n,k-1)|+a2(n,k-1)<1 (37)
a2(n,k-1)>-1 or |a1(n,k-1)|<-2 (38)
Configuration of Evaluation Function Value Calculating Unit
[0204] The configuration of the evaluation function value
calculating unit 512 will be explained. The evaluation function
value calculating unit 512 calculates the evaluation function value
J such as that shown in the following formula (39), based on the
square sums of deviation between the predicted ammonia
concentration PREDNH3.sub.EXS calculated by the predictor 511 and
the corrected target ammonia concentration
DNH3.sub.CONS.sub.--.sub.TRGT.sub.--.sub.MOD calculated by the
target value correction unit 522.
J ( n ) = i = 1 np ( PRENH 3 EXS ( n + MmiMie i ) - DNH 3 CONS_TRGT
_MOD ( k + i ) ) 2 ( 39 ) ##EQU00013##
[0205] FIG. 6 presents graphs showing a relationship between the
evaluation function value J and the urea injection amount
difference DG.sub.UREA.
[0206] In FIG. 6, the top graph shows the relationship between the
ammonia concentration and time, and the bottom graph shows the
relationship between the urea injection amount and time. FIG. 6
shows a case in which predicted ammonia concentrations
PREDNH3.sub.EXS(k+1) to PREDNH3.sub.EXS(k+np) on and after the
present time are calculated based on the detected ammonia
concentration DNH3.sub.CONS and urea injection amount difference
DG.sub.UREA until the present time, with the time k shown by the
double circle as the present time.
[0207] As shown in FIG. 6, the virtual target value
DNH3.sub.CONS.sub.--.sub.TRGT.sub.--.sub.F shows a first order lag
in response to the target ammonia concentration difference
DNH3.sub.CONS.sub.--.sub.TRGT set step-wise. Furthermore, as
explained above, the corrected target ammonia concentration
DNH3.sub.CONS.sub.--.sub.TRGT.sub.--.sub.MOD(k+i) is set between
the predicted ammonia concentration PREDNH.sub.EXS(k+i), which is a
future predicted value of the detected ammonia concentration
difference DNH3.sub.CONS shown by a solid line, and the target
ammonia concentration difference DNH3.sub.CONS.sub.--.sub.TRGT
(k+i).
[0208] Herein, the predicted ammonia concentrations
PREDNH3.sub.EXS(k+1) to PREDNH3.sub.EXS(k+np) are calculated under
conditions in which the urea injection amount differences
DG.sub.UREA(k+1) to DG.sub.UREA(k+np-1) at this time and after are
made constant, as shown in the above formula (19).
[0209] In addition, the evaluation function value J becomes the
square sum of the deviation between these predicted ammonia
concentration PREDNH3.sub.EXS and corrected target ammonia
concentrations DNH3.sub.CONS.sub.--.sub.TRGT.sub.--.sub.MOD, as
shown in the above formula (39). Therefore, by determining the urea
injection amount difference DG.sub.UREA(k) so as to make such an
evaluation function value J a minimum, the urea injection amount
difference DG.sub.UREA can be determined so that the region
indicated in hatching in FIG. 6 becomes narrower, i.e. the detected
ammonia concentration difference DNH3.sub.CONS quickly matches the
target ammonia concentration difference
DNH3.sub.CONS.sub.--.sub.TRGT.
[0210] Referring back to FIG. 5, although the evaluation function
value J is defined by the above formula (39) in the present
embodiment, this definition for the evaluation function value J may
be altered as appropriate depending on the use.
[0211] More specifically, in a case of the prediction time
extending to reduce the computational load, for example, it may be
defined by thinning the square sum of deviation by a predetermined
interval, as shown in the following formula (40). Herein, the
constant q is an integer specifying the thinning interval, and is
set as the denominator of the number of prediction steps np.
J ( n ) = i = 1 np / q ( PRENH 3 EXS ( n + MmiMieqi ) - DNH 3
CONS_TRGT _MOD ( k + qi ) ) 2 ( 40 ) ##EQU00014##
[0212] Alternatively, in a case of further reducing the
computational lbad, for example, it may be defined by the square of
deviation after the prediction step number np, as shown in the
following formula (41).
J(n)=(PRENH3.sub.EXS(n+MmiMienp)-DNH3.sub.CONS.sub.--.sub.TRGT.sub.--.su-
b.MOD(k+np)).sup.2 (41)
Configuration of Extremum Searching Optimization Unit
[0213] The configuration of the extremum searching optimization
unit 513 will be explained.
[0214] The extremum searching optimization unit 513 is configured
to include an amplifier 551, a washout filter 552, a moving average
value calculating unit 553, a sliding mode controller 554, a
multiplier 555, an adder 556, and a reference signal output unit
557.
[0215] This extremum searching optimization unit 513 calculates the
optimum injection amount DG.sub.UREA.sub.--.sub.OPT and the search
input DG.sub.UREA.sub.--.sub.EXS according to the sequence shown in
the below formulas (42) to (51), so that the evaluation function
value J thus calculated becomes a minimum.
[0216] The amplifier 551 inverts the sign of the evaluation
function value J by multiplying the evaluation function value J
thus calculated by the evaluation function value calculating unit
512 by "-1" and redefines this as evaluation function value J', as
shown in the below formula (42). In other words, for convenience of
operation, this extremum searching optimization unit 513 searches
for the maximum value of the evaluation function value J' instead
of searching for the minimum value of the evaluation function value
J.
J'(n)=-J(n) (42)
[0217] The washout filter 552 calculates a high-pass filtering
value J.sub.WF of this evaluation function value J' by conducting
simplified filter processing, such as that shown in the below
formula (43), on the evaluation function value J'. It should be
noted that the passband of the filter in this filter processing is
set so as to include the frequency F.sub.REF of the reference
signal S.sub.REF (refer to formula (45) described later).
J.sub.WF(n)=0.5J'(n)-0.5J'(n-1) (43)
[0218] The multiplier 555 multiplies the reference signal S.sub.REF
output from the reference signal output unit 557 by the high-pass
filtering value J.sub.WF of the evaluation function value J' to
calculate the product C.sub.R, as shown in the following formula
(44).
C.sub.R(n)=J.sub.WF(n)S.sub.REF(n) (44)
[0219] Herein, the reference signal output unit 557 outputs a sine
wave of amplitude A.sub.REF and frequency F.sub.REF, such as that
shown in the following formula (45), as the reference signal
S.sub.REF. It should be noted that the reference signal is not
limited to such a sine wave, and may adopt any periodic function
such as a triangle wave, trapezoid wave, and synthetic wave.
S.sub.REF(n)=A.sub.REF sin(2.pi.F.sub.REFn.DELTA.Te) (45)
[0220] The moving average value calculating unit 553 calculates the
moving average value C.sub.R.sub.--.sub.AVE of the product C.sub.R,
as shown in the following formula (46). Herein, N.sub.AVE is the
moving average tap number.
C R_AVE ( n ) = i = 0 Nave C R ( n - i ) ( 46 ) ##EQU00015##
[0221] FIG. 7 presents graphs showing a correlation between the
evaluation function value J' and the moving average value
C.sub.R.sub.--.sub.AVE relative to the optimum injection amount
DG.sub.UREA.sub.--.sub.OPT. More specifically, FIG. 7 presents
graphs showing the change in the evaluation function value J' and
the moving average value C.sub.R.sub.--.sub.AVE when causing the
optimum injection amount DG.sub.UREA.sub.--.sub.OPT to change.
[0222] As shown in FIG. 7, the moving average value
C.sub.R.sub.--.sub.AVE becomes "0" with the optimum injection
amount DG.sub.UREA.sub.--.sub.OPT shown by the star at which the
evaluation function value J' is a maximum. In addition, the moving
average value C.sub.R.sub.--.sub.AVE is a monotone increasing
function relative to the optimum injection amount
DG.sub.UREA.sub.--.sub.OPT.
[0223] Based on this fact, the extremum searching optimization unit
513 uses the moving average value C.sub.R.sub.--.sub.AVE shown in
the above formula (46) as a correlation function, and determines
the optimum injection amount DG.sub.UREA.sub.--.sub.OPT such that
this moving average value C.sub.R.sub.--.sub.AVE becomes "0".
[0224] Referring back to FIG. 5, the moving average tap number
N.sub.AVE of the above formula (46) is defined by the following
formula (47). Herein, supposing j is an integer, the time interval
of the moving average segment, i.e. N.sub.AVE.DELTA.Te, of the
moving average value C.sub.R.sub.--.sub.AVE becomes the integer j
times the period of the reference signal S.sub.REF.
N AVE = j 1 F REF .DELTA. Te ( 47 ) ##EQU00016##
[0225] As shown in FIG. 5, a loop in which a periodic signal
containing the reference signal S.sub.REF recurs is formed between
the virtual plant 515 and the extremum searching optimization unit
513. When such a periodic reference signal S.sub.REF recurs, the
parameters in this loop become oscillatory, and these parameters
may diverge in the worst case.
[0226] Consequently, in the present embodiment, by defining the
moving average tap number N.sub.AVE as shown in the above formula
(47), the periodic component of the reference signal S.sub.REF is
eliminated from the moving average value C.sub.R.sub.--.sub.AVE,
and the periodic component of the reference signal S.sub.REF is
prevented from recurring in the aforementioned loop. In this way,
the maximum value of the evaluation function value J' can be stably
searched.
[0227] The sliding mode controller 554 calculates the optimum
injection amount DG.sub.UREA.sub.--.sub.OPT such that the moving
average value C.sub.R.sub.--.sub.AVE becomes "0", based on an
algorithm of sliding mode control. Herein, sliding mode control is
an extension of so-called response-adaptive control that enables
designation of a convergence rate of a controlled amount, and is
control that enables the following rate of the controlled amount to
the target value and the convergence rate of the controlled amount
in a case of noise having been applied to be individually
designated. More specifically, it calculates the optimum injection
amount DG.sub.UREA.sub.--.sub.OPT based on the following formulas
(48) to (51).
DG.sub.UREA.sub.--.sub.OPT(n)=DG.sub.UREA.sub.--.sub.RCH(n)+DG.sub.UREA.-
sub.--.sub.ADP(n) (48)
DG.sub.UREA.sub.--.sub.RCH(n)=K.sub.RCH.sigma.(n) (49)
DG UREA_ADP ( n ) = K ADP i = 1 n .sigma. ( i ) ( 50 ) ##EQU00017##
.sigma.(n)=C.sub.R.sub.--.sub.AVE(n)+SC.sub.R.sub.--.sub.AVE(n-1)
(51)
[0228] As shown in the above formula (48), the optimum injection
amount DG.sub.UREA.sub.--.sub.OPT is calculated from the sum of the
reaching-law input DG.sub.UREA.sub.--.sub.RCH shown in the above
formula (49) and the adaptive-law input DG.sub.UREA.sub.--.sub.APP
shown in the above formula (50).
[0229] The reaching-law input DG.sub.UREA.sub.--.sub.RCH is an
input for placing the moving average value C.sub.R.sub.--.sub.AVE
on a switching line described later, and is calculated by
multiplying a predetermined reaching-law control gain K.sub.RCH by
a switching function o shown in the above formula (51)
[0230] The adaptive-law input DG.sub.UREA.sub.--.sub.ADP is an
input for suppressing the influence of modeling error and noise and
places the moving average value C.sub.R.sub.--.sub.AVE on the
switching line described later, and is calculated by multiplying a
predetermined adaptive-law control gain K.sub.ADP by the value of
an integral of the switching function .sigma..
[0231] The switching function .sigma. is calculated by the sum of
the moving average value C.sub.R.sub.--.sub.AVE(n) of the current
control time and the product of multiplying a predetermined
switching function setting parameter S by the moving average value
C.sub.R.sub.--.sub.AVE(n-1) of the previous control time, as shown
in the above formula (51).
[0232] Herein, the relationship between the switching function
setting parameter S and the convergence rate of
C.sub.R.sub.--.sub.AVE will be explained.
[0233] As shown in the above formula (51), the combination of the
moving average value C.sub.R.sub.--.sub.AVE(n) and
C.sub.R.sub.--.sub.AVE(n-1) satisfying the switching function
.sigma.(n)=0 within a phase plane defining the horizontal axis as
the moving average value C.sub.R.sub.--.sub.AVE(n-1) of the
previous control time and the vertical axis as the moving average
value C.sub.R.sub.--.sub.AVE(n) of the current control time is a
straight line with a slope of -S. In particular, this straight line
is called the switching line. In addition, on this switching line,
since C.sub.R.sub.--.sub.AVE(n-1)>C.sub.R.sub.--.sub.AVE(n) by
setting -S to a value larger than 0 and less than 1, the moving
average value C.sub.R.sub.--.sub.AVE(n) converges to 0. The sliding
mode control is control focused on the behavior of the moving
average value C.sub.R.sub.--.sub.AVE(n) on this switching line.
[0234] In other words, by performing control so that the
combination of the moving average value C.sub.R.sub.--.sub.AVE(n)
of the current control time and the moving average value
C.sub.R.sub.--.sub.AVE(n-1) of the previous control time falls on
this switching line, control that is robust against noise and
modeling error is realized, and the moving average value
C.sub.R.sub.--.sub.AVE can be made to converge to "0" while
suppressing overshoot and oscillatory behavior.
[0235] In particular, since the evaluation function value J changes
depending on the detected ammonia concentration difference
DNH3.sub.CONS and the virtual target value
DNH3.sub.CONS.sub.--.sub.TRGT.sub.--.sub.F, the change thereof
during a transition becomes large, a result of which the moving
average value C.sub.R.sub.--.sub.AVE easily displays oscillatory
behavior. By using sliding mode control in such a system,
oscillation occurring in the moving average value
C.sub.R.sub.--.sub.AVE can be effectively suppressed.
[0236] It should be noted that, although the optimum injection
amount DG.sub.UREA.sub.--.sub.OPT has been calculated so that the
moving average value C.sub.R.sub.--.sub.AVE converges to "0" by way
of an algorithm of sliding mode control in the present embodiment,
it is not limited thereto, and PID control may be used. However, if
using PID control, the moving average value C.sub.R.sub.--.sub.AVE
will easily display oscillatory behavior compared to a case of
using sliding mode control. Therefore, the case of using PID
control requires establishing so that the oscillatory behavior of
the moving average value C.sub.R.sub.--.sub.AVE is within an
acceptable range.
[0237] In addition, the optimum injection amount
DG.sub.UREA.sub.--.sub.OPT calculated in the above way is output as
the FB injection amount G.sub.UREA.sub.--.sub.FR of the feedback
controller 51.
[0238] The adder 556 calculates the search input
DG.sub.UREA.sub.--.sub.EXS by adding the reference signal S.sub.REF
to the optimum injection amount DG.sub.UREA.sub.--.sub.OPT not
containing the component of the reference signal S.sub.REF, as
shown in the following formula (52).
DG.sub.UREA.sub.--.sub.EXS(n)=DG.sub.UREA.sub.--.sub.OPT(n)+S.sub.REF(n)
(52)
[0239] With the extremum searching optimization unit 513 configured
in the above way, the update period .DELTA.Te of the optimum
injection amount DG.sub.UREA.sub.--.sub.OPT and the search input
DG.sub.UREA.sub.--.sub.EXS is made shorter than the update period
.DELTA.Ti of the urea injection amount G.sub.UREA. With this,
oscillations when optimizing the optimum injection amount
DG.sub.UREA.sub.--.sub.OPT and the search input
DG.sub.UREA.sub.--.sub.EXS by the extremum searching optimization
unit 513 can be prevented from appearing as oscillations of the
urea injection amount G.sub.UREA actually input to the exhaust
purification device 2. Therefore, it is possible to stably control
the exhaust purification device 2.
[0240] Control using the extremum searching optimization unit 513
of the present embodiment configured in the above way and control
using a conventional extremum searching optimization unit will be
compared.
[0241] FIG. 8 is a block diagram showing a configuration between a
conventional extremum searching optimization unit 513A and a
control object 2A.
[0242] The conventional extremum searching optimization unit 513A
inputs a control input value containing the component of the
periodic reference signal S.sub.REF to a direct control object.
Furthermore, optimization of the control input value is performed
based on the control output value as a response to the input
containing this reference signal S.sub.REF so that the control
input value excluding the reference signal S.sub.REF matches a
predetermined target value.
[0243] On the other hand, the extremum searching optimization unit
513 of the present embodiment optimizes the control input value
DG.sub.UREA.sub.--.sub.OPT so as to minimize the evaluation
function value J. As a result, it differs from the aforementioned
conventional extremum searching optimization unit 513A in that it
does not directly input the search input DG.sub.UREA.sub.--.sub.EXS
containing the reference signal S.sub.REF to the exhaust
purification device (refer to FIG. 4) as a control object, but
rather sets up the virtual plant 515 and inputs to this virtual
plant 515, as shown in FIG. 5. Furthermore, it optimizes the
control input values DG.sub.UREA.sub.--.sub.OPT and
DG.sub.UREA.sub.--.sub.EXS so that the output J of this virtual
plant becomes a minimum. This point differs greatly from the case
of using the conventional extremum searching optimization unit
513A.
[0244] Referring back to FIG. 4, the reference injection amount
G.sub.UREA.sub.--.sub.BS will be explained in detail next.
[0245] In the present embodiment, the FF injection amount
G.sub.UREA.sub.--.sub.FF is used as the reference injection amount
G.sub.UREA.sub.--.sub.BS. This FF injection amount
G.sub.UREA.sub.--.sub.FF is a control input value for controlling
so that the NOx reduction rate of the selective reduction catalyst
maintains the maximum value, and is determined based on the amount
of NOx in the exhaust, which changes according to the operating
state of the engine.
[0246] More specifically, this FF injection amount
G.sub.UREA.sub.--.sub.FF is calculated based on the output value of
the NOx sensor 28 (refer to FIG. 1), i.e. detected NOx
concentration NOX.sub.CONS, as shown in the following formulas (53)
and (54).
G.sub.UREA.sub.--.sub.FF(m)=K.sub.NOX.sub.--.sub.NH3NOX.sub.CONS(m-.delt-
a.1)G.sub.EX(m-.delta.2) (53)
G.sub.EX(m)=K.sub.FUEL.sub.--.sub.EXG.sub.FUEL(m)NE(m)/60 (54)
[0247] Herein, .delta.1 in the above formulas (53) and (54)
indicates the time until the exhaust reaches the selective
reduction catalyst from the detection portion of the NOx sensor. In
addition, .delta.2 indicates the time until the exhaust reaches the
selective reduction catalyst from the engine. G.sub.EX indicates an
estimated value for the volume of exhaust. K.sub.NOX.sub.--.sub.NH3
indicates a coefficient for calculating the urea injection amount
necessary for reducing NOx. K.sub.FUEL.sub.--.sub.EX indicates a
coefficient for converting from fuel injection amount to volume of
exhaust. NE indicates the revolution speed of the engine.
[0248] It should be noted that, in a case of not using the output
value of the NOx sensor, the FF injection amount
G.sub.UREA.sub.--.sub.FF is determined from a map search, based on
the revolution speed NE of the engine and a load parameter TRQ
representing the load of the engine as the operating state of the
engine, for example.
[0249] FIG. 9 is a graph showing an example of a control map for
determining the FF injection amount G.sub.UREA.sub.--.sub.FF.
[0250] As shown in FIG. 9, in the control map, the FF injection
amount G.sub.UREA.sub.--.sub.FF is determined to be a larger value
accompanying the revolution speed NE of the engine or the load
parameter TRQ increasing.
[0251] This is because, with a larger load parameter TRQ of the
engine, the NOx emission amount increases from the combustion
temperature of the air-fuel mixture rising, and with a higher
revolution speed NE of the engine, the NOx emission amount per unit
time increases.
[0252] Although the FF injection amount G.sub.UREA.sub.--.sub.FF
determined in the aforementioned way has been used as the reference
injection amount G.sub.UREA.sub.--.sub.BS in the present
embodiment, it is not limited thereto. For example, as shown in the
following formulas (55) to (57), the reference injection amount
G.sub.UREA.sub.--.sub.BS may be determined so as to be adapted to
degradation and variation in the selective reduction catalysts and
variation in the NOx emission characteristics of engines.
G.sub.UREA.sub.--.sub.BS(m)=G.sub.UREA.sub.--.sub.BS.sub.--.sub.INI+DG.s-
ub.UREA.sub.--.sub.BS(m) (55)
DG UREA_BS ( m ) = KP BS E BS ( m ) + KI BS i = 0 m E BS ( i ) ( 56
) E BS ( m ) = { - DG UREA_FB ( m ) or DNH 3 CONS ( m ) - DNH 3
CONS_TRGT _F ( m ) ( 57 ) ##EQU00018##
[0253] Herein, the G.sub.UREA.sub.--.sub.BS.sub.--.sub.INI in the
above formulas (55) to (57) indicates the initial value of
G.sub.UREA.sub.--.sub.BS. KP.sub.BS and KI.sub.BS respectively
indicate adaptive gains. DG.sub.UREA.sub.--.sub.FB indicates the
down sampling value by the period .DELTA.Ti of
DG.sub.UREA.sub.--.sub.OPT, which is updated at the period
.DELTA.Te.
[0254] Next, urea injection control processing executed by the ECU
will be explained while referring to FIG. 10.
[0255] FIG. 10 is a flowchart showing a sequence of urea injection
control processing executed by the ECU.
[0256] In Step S1, it is distinguished whether a urea fault flag
F.sub.UREANG is "1". This urea fault flag F.sub.UREANG is set to
"1" when it is determined that the urea injection device has failed
in determination processing, which is not illustrated, and is set
to "0" at times except for this. In a case of this determination
being YES, Step S9 is advanced to, and after the urea injection
amount G.sub.UREA has been set to "0", this processing ends. In a
case of this determination being NO, Step S2 is advanced to.
[0257] In Step S2, it is distinguished whether a catalyst
degradation flag F.sub.SCRNG is "1". This catalyst degradation flag
F.sub.sGRNG is set to "1" when it is determined that either the
first selective reduction catalyst and the second selective
reduction catalyst has failed in the determination processing,
which is not illustrated, and is set to "0" at times except for
this. In a case of this determination being YES, Step S9 is
advanced to, and after the urea injection amount G.sub.UREA has
been set to "0", this processing ends. In a case of this
determination being NO, Step S3 is advanced to.
[0258] In Step S3, it is determined whether a urea remaining amount
Q.sub.UREA is less than a predetermined value Q.sub.REF. This urea
remaining amount Q.sub.UREA indicates the remaining amount of urea
water in the urea tank, and is calculated based on the output of
the urea level sensor. In a case of this determination being YES,
Step S4 is advanced to, and in a case of being NO, Step S5 is
advanced to.
[0259] In Step S4, a urea remaining amount warning light is
illuminated, Step S9 is advanced to, and after the urea injection
amount G.sub.UREA has been set to "0", this processing ends.
[0260] In Step S5, it is determined whether a catalyst warm-up
timer value T.sub.MAST is larger than a predetermined value
T.sub.MLMT. This catalyst warm-up timer value T.sub.MAST is a value
keeping the warm-up time of the selective reduction catalyst after
engine startup. In a case of this determination being YES, Step S6
is advanced to. In a case of this determination being NO, Step S9
is advanced to, and after the urea injection amount G.sub.UREA has
been set to "0", this processing ends.
[0261] In Step S6, it is distinguished whether a sensor fault flag
F.sub.SENNG is "0". This sensor fault flag F.sub.SENNG is set to
"1" when it is determined that the ammonia sensor or the catalyst
temperature sensor has failed in the determination processing,
which is not illustrated, and is set to "0" at times except for
this. In a case of this determination being YES, Step S7 is
advanced to. In a case of this determination being NO, Step S9 is
advanced to, and after the urea injection amount G.sub.UREA has
UREA been set to "0", this processing ends.
[0262] In Step S7, it is distinguished whether an ammonia sensor
activity flag F.sub.NH3ACT is 1. This ammonia sensor activity flag
F.sub.NH3ACT is set to "1" when it is determined that the ammonia
sensor has reached an active state in determination processing,
which is not illustrated, and is set to "0" at times except for
this. In a case of this determination being YES, Step S8 is
advanced to. In a case of this determination being NO, Step S9 is
advanced to, and after the urea injection amount G.sub.UREA has
been set to "0", this processing ends.
[0263] In Step S8, it is distinguished whether the temperature
T.sub.SCR of the first selective reduction catalyst is higher than
a predetermined value T.sub.SCR.sub.--.sub.ACT. In a case of this
determination being YES, it is determined that the first selective
reduction catalyst has been activated, and Step S10 is advanced to.
In a case of this determination being NO, it is determined that the
first selective reduction catalyst has not been activated yet and
that urea injection should be stopped, Step S9 is advanced to, and
after the urea injection amount G.sub.UREA has been set to "0",
this processing ends.
[0264] In Step S10, the virtual target value
DNH3.sub.CONS.sub.--.sub.TRGT.sub.--.sub.F is calculated based on
the formulas (5) and (6) by the aforementioned target value filter
521, and Step S11 is advanced to.
[0265] In Step S11, the corrected target ammonia concentration
DNH3.sub.CONS.sub.--.sub.TRGT.sub.--.sub.MOD is calculated based on
the formulas (7) and (8) by the aforementioned target value
correction unit 522, and Step S12 is advanced to.
[0266] In Step S12, the model parameters a1, a2, b1 and b2 are
calculated based on the formulas (9) to (15) by the aforementioned
identifier 53, and Step S13 is advanced to.
[0267] In Step S13, the predicted ammonia concentration
PREDNH3.sub.EXS is calculated based on the formulas (27) to (35) by
the aforementioned predictor 511, and Step S14 is advanced to.
[0268] In Step S14, the evaluation function value J is calculated
based on the formula (39) by the aforementioned evaluation function
value calculating unit 512, and Step S15 is advanced to.
[0269] In Step S15, the optimum injection amount
DG.sub.UREA.sub.--.sub.OPT (FB injection amount
G.sub.UREA.sub.--.sub.FB) and search input
DG.sub.UREA.sub.--.sub.EXS are calculated based on the formulas
(42) to (51) by the aforementioned extremum searching optimization
unit 513, and Step S16 is advanced to.
[0270] In Step S16, the FF injection amount
G.sub.UREA.sub.--.sub.FF (reference injection amount
G.sub.UREA.sub.--.sub.BS) is calculated based on the formulas (53)
and (54), and Step S17 is advanced to.
[0271] In Step S17, the urea injection amount G.sub.UREA is
calculated based on formula (1) by the adder 54, and this
processing ends.
[0272] Next, the results of an example of control by way of the ECU
of the present embodiment will be explained while referring to FIG.
11.
[0273] FIG. 11 presents graphs showing relationships between the
detected ammonia concentration NH3.sub.CONS, the evaluation
function value J', and the urea injection amount G.sub.UREA in a
case of performing urea injection control by way of the ECU of the
present embodiment. It should be noted that this example shown in
FIG. 11 is an example of performing control under the same
conditions as the aforementioned example shown in FIG. 3. In other
words, after the target ammonia concentration
NH3.sub.CONS.sub.--.sub.TRGT is made to increase step-wise from "0"
to a predetermined value at the time t.sub.0, modeling error and
noise is imparted at time t.sub.A, time t.sub.B, and time
t.sub.C.
[0274] In the aforementioned example of conventional control shown
in FIG. 3, immediately after the target ammonia concentration
NH3.sub.CONS.sub.--.sub.TRGT is made to change step-wise at time
t.sub.0, and immediately after noise or modeling error is applied
at times t.sub.A, t.sub.B, and t.sub.C, the detected ammonia
concentration NH3.sub.CONS shows overshoot and oscillatory behavior
relative to the target ammonia concentration
NH3.sub.CONS.sub.--.sub.TRGT. In addition, the urea injection
amount G.sub.UREA is decreased more than necessary in response to
such overshoot and oscillatory behavior of the detected ammonia
concentration NH3.sub.CONS.
[0275] In contrast, it has been confirmed that, in the example of
control of the present embodiment shown in FIG. 11, such overshoot
and oscillatory behavior of the detected ammonia concentration
NH3.sub.CONS when the target ammonia concentration
NH3.sub.CONS.sub.--.sub.TRGT is made to change, and the overshoot
and oscillatory behavior of the detected ammonia concentration
NH3.sub.CONS immediately after noise and modeling error is applied
are suppressed. In addition, it has also been confirmed that, by
suppressing such overshoot and oscillatory behavior of the detected
ammonia concentration NH3.sub.CONS, the urea injection amount
G.sub.UREA is not decreased more than necessary.
[0276] Moreover, it could be confirmed that the ECU of the present
embodiment can reduce the computational load on the order of
several to tens of times compared to model prediction control by
convention evaluation norm, by calculating the urea injection
amount G.sub.UREA by way of the aforementioned such virtual plant
515 and extremum searching algorithm of the extremum searching
optimization unit 513. Therefore, even in a case in which a
computing device having high computing power cannot be used due to
being in an adverse environment such as high temperature, high
humidity, high vibration, or dust, it is possible to perform
control with high precision using an ECU of a simple
configuration.
Second Embodiment
[0277] Next, a second embodiment of the present invention will be
explained while referring to the drawings.
[0278] FIG. 12 is a schematic diagram showing configurations of an
ECU 7 as a control device and a vehicle 6 as a plant controlled by
this ECU 6 according to the second embodiment of the present
invention.
[0279] This vehicle 6 includes drive wheels 61 and undriven wheels
62, and an unillustrated engine that generates torque for
rotationally driving the drive wheels 61.
[0280] The ECU 7 calculates the target value of torque (hereinafter
referred to as "target torque") TRQ of the engine in the vehicle 6,
and inputs this target torque TRQ to the vehicle 6 as a control
input value. In addition, a drive-wheel speed sensor (not
illustrated) that detects the rotational speed (hereinafter
referred to as "drive-wheel speed") W.sub.S.sub.--.sub.ACT of the
drive wheels 61 that are rotationally driven is provided to the
vehicle 6, and the detected value W.sub.S.sub.--.sub.ACT of this
drive-wheel speed sensor is output to the ECU 7 as a control output
value.
[0281] For example, when excessive torque acts on the drive wheels
61 such as during acceleration of the vehicle 6, the drive wheels
61 will excessively spin relative to the undriven wheels 62, and
the stability and acceleration of the vehicle will decline
considerably. This ECU 7 prevents such excessive spinning of the
drive wheels 61 by calculating the target torque TRQ so that the
drive-wheel speed W.sub.S.sub.--.sub.ACT matches a predetermined
target speed. In other words, a so-called traction control system
is configured by this vehicle 6 and ECU 7. Hereinafter,
configurations of modules of the ECU 7 in this traction control
system will be explained.
[0282] The modules shown in FIG. 12 are configured to include a
feedback controller 71, feedforward controller 72, target value
calculating portion 73, target value correction portion 74,
identifier 75, and a plurality of adders 76 and 77.
[0283] The target value calculating portion 73 calculates a target
value of the drive-wheel speed W.sub.S.sub.--.sub.ACT (hereinafter
referred to as "target wheel speed") W.sub.S.sub.--.sub.CMD based
on the rotational speed (hereinafter referred to as "undriven wheel
speed") W.sub.S.sub.--.sub.REF of the undriven wheels 62 output
from the vehicle 6, as explained in detail later.
[0284] As explained in detail later, the feedback controller 71
includes a target value correction unit 742, predictor 712,
evaluation function value calculating unit 713, and extremum
searching optimization unit 714, and calculates a feedback torque
(hereinafter referred to as "FB torque") DTRQ.sub.FB, so that the
drive-wheel speed W.sub.S.sub.--.sub.ACT matches the target wheel
speed W.sub.S.sub.--.sub.CMD.
[0285] The feedforward controller 72 calculates a feedforward
torque (hereinafter referred to as "FF torque") TRQ.sub.FF based on
the revolution speed of the engine (hereinafter referred to as
"engine revolution speed") and a depression amount AP of the
accelerator pedal output from the vehicle 6, as explained in detail
later.
[0286] The adder 76 calculates the target torque TRQ by adding an
FB torque DTRQ.sub.FB calculated by the feedback controller 71 and
the FF torque TRQ.sub.FF calculated by the feedforward controller
72, as shown in the following formula (58), and inputs this to the
vehicle 6 as a control input value.
TRQ=TRQ.sub.FF+DTRQ.sub.FB (58)
[0287] The target value correction portion 74 calculates a
corrected target wheel speed W.sub.S.sub.--.sub.CMD.sub.--.sub.MOD
by conducting a predetermined operation on the target wheel speed
W.sub.S.sub.--.sub.CMD in order to prevent overshoot and
oscillatory behavior of the drive wheel speed
W.sub.S.sub.--.sub.ACT relative to the target wheel speed
W.sub.S.sub.--.sub.CMD, as explained in detailed later.
[0288] The identifier 75 identifies a plurality of model parameters
a1', a2', b1' and b2' used by the predictor 712 of the feedback
controller 71, as explained in detail later.
[0289] It should be noted that an operation by the respective
modules shown in FIG. 12 every different control cycle.
Hereinafter, the control cycle of the adders 76 and 77, i.e. the
update period of a search input TRQ.sub.EXS described later and the
target torque TRQ, is set to .DELTA.Tc. In addition, the control
cycle of the identifier 75 and the feedforward controller 72 is
also set to .DELTA.Tc. In other words, the update periods of the
model parameters a1', a2', b1' and b2' and the FF torque TRQ.sub.FF
are also set to .DELTA.Tc. It should be noted that this update
period .DELTA.Tc is set to approximately 0.002 seconds in the
present embodiment, for example.
[0290] Hereinafter, the configuration of each module will be
explained in detail.
Configuration of Target Value Calculating Portion
[0291] The configuration of the target value calculating portion 73
will be explained.
[0292] The target value calculating portion 73 calculates the
target wheel speed W.sub.S.sub.--.sub.CMD of the drive wheel speed
W.sub.S.sub.--.sub.ACT based on the undriven wheel speed
W.sub.S.sub.--.sub.REF output from the vehicle 6. More
specifically, the target wheel speed W.sub.S.sub.--.sub.CMD is
calculated by adding a predetermined wheel slip offset amount
OPT.sub.SLIP to the undriven wheel speed W.sub.S.sub.--.sub.REF, as
shown in the following formula (59). In other words, this wheel
slip offset amount OPT.sub.SLIP corresponds to an allowable slip
amount between the drive wheels 61 and the undriven wheels 62.
W.sub.S.sub.--.sub.CMD(m)=W.sub.S.sub.--.sub.REF(m)+OPT.sub.SLIP
(59)
[0293] Herein, the notation "m" is a notation indicating
discretized time, and indicates being data detected or calculated
every update period .DELTA.Tc of the aforementioned target torque
TRQ.
[0294] In addition, although a predetermined constant is set as the
wheel slip offset amount OPT.sub.SLIP in the present embodiment, it
is not limited thereto. For example, it may be determined based on
parameters such as an estimated value of the road surface
coefficient of friction, a detected value of a steering wheel
angle, a detected value of a yaw rate sensor, and a detected value
of a car body slip angle.
Configuration of Target Value Correction Portion
[0295] The configuration of the target value correction portion 74
will be explained.
[0296] The target value correction portion 74 is configured to
include a target value filter 741 and a target value correction
unit 742, and calculates a corrected target wheel speed
W.sub.S.sub.--.sub.CMD.sub.--.sub.MOD by the sequence shown in the
following formulas (60) to (63).
[0297] The target value filter 741 is a so-called first-order lag
filter. In other words, the target value filter 741 calculates a
virtual target value W.sub.S.sub.--.sub.CMD.sub.--.sub.F by
performing the operation shown in the following formulas (60) and
(61) on the target wheel speed W.sub.S.sub.--.sub.CMD
calculated.
W.sub.S.sub.--.sub.CMD(k+i)=W.sub.S.sub.--.sub.CMD(k)
(i=0.about.mp) (60)
W S_CMD _F ( k + i ) = - POLE F W S_CMD _F ( k + i - 1 ) + ( 1 +
POLE F ' ) W S_CMD ( k + i ) ( i = 0 ~ mp ) ( 61 ) ##EQU00019##
[0298] Herein, the notation "k" is notation indicating discretized
time, and indicates being data detected or calculated every model
sampling period .DELTA.Tm described later.
[0299] In addition, POLE.sub.F' is a tracking characteristic
designating parameter for designating the tracking characteristic
of the drive wheel speed W.sub.S.sub.--.sub.ACT relative to the
target wheel speed W.sub.S.sub.--.sub.CMD, and POLE.sub.F' is set
to be greater than -1 and less than 0. In addition, mp in the above
formulas (60) and (61) is a positive integer, and indicates a
prediction step number (step number for which .DELTA.Tm is set as
the control cycle) in a predictor 712.
[0300] The target value correction unit 742 calculates the
corrected target wheel speed W.sub.S.sub.--.sub.CMD.sub.--.sub.MOD
according to the response delay characteristic of the drive wheel
speed W.sub.S.sub.--.sub.ACT as shown in the following formulas
(62) and (63), in order to prevent overshoot and oscillatory
behavior of the drive wheel speed W.sub.S.sub.--.sub.ACT relative
to the target wheel speed W.sub.S.sub.--.sub.CMD.
[0301] More specifically, this target value correction unit 742
first calculates deviation Ew between the drive wheel speed
W.sub.S.sub.--.sub.ACT and the virtual target value
W.sub.S.sub.--.sub.CMD.sub.--.sub.F, as shown in the following
formula (62).
Ew(k)=W.sub.S.sub.--.sub.ACT(k)-W.sub.S.sub.--.sub.CMD.sub.--.sub.F(k)
(62)
[0302] Next, the corrected target wheel speed
W.sub.S.sub.--.sub.CMD.sub.--.sub.MOD is calculated by adding a
damping term proportional to the deviation Ew to the virtual target
value W.sub.S.sub.--.sub.CMD.sub.--.sub.F, as shown in the
following formula (63).
W.sub.S.sub.--.sub.CMD.sub.--.sub.MOD(k+i)=W.sub.S.sub.--.sub.CMD.sub.---
.sub.F(k)+(1+POLE.sub.E').sup.iEw(k) (63)
[0303] Herein, POLE.sub.E' contained in the damping term of the
above formula (63) is a deviation convergence characteristic
designating parameter for designating the convergence
characteristic of the deviation Ew, and POLE.sub.E' is set to be
greater than -1 and less than 0. In the above way, the corrected
target wheel speed W.sub.S.sub.--.sub.CMD.sub.--.sub.MOD(k+1),
which is a target value of a future time (k+1), is set between the
target wheel speed W.sub.S.sub.--.sub.CMD(k+1) at a future time and
a future predicted value of the drive wheel speed
W.sub.S.sub.--.sub.ACT (predicted drive wheel speed
PREW.sub.S.sub.--.sub.EXS (k+1) described later), taking into
account of the response delay characteristic of the drive wheel
speed W.sub.S.sub.--.sub.ACT.
Configuration of Identifier
[0304] The configuration of the identifier 75 will be
explained.
[0305] The identifier 75 identifies the plurality of model
parameters a1', a2', b1' and b2' of a plant model used by the
predictor 712 by way of the recursive identification algorithms
(generalized recursive least-squares method algorithms) shown in
the following formulas (64) to (70). More specifically, a current
control value .theta.(m) of a model parameter vector is calculated
based on a latest control value .xi.' of processing object data and
a previous control value .theta.(m-1) of the model parameter
vector.
[0306] The .theta.'(m) and .xi.(m) are respectively defined as
4-component vectors, as shown in the following formulas (64) and
(65).
.theta.'.sup.T'(m)=[a1'(m)a2'(m)b1'(m)b2'(m)] (64)
.zeta.'.sup.T(m)=[W.sub.S.sub.--.sub.ACT(m-Mmi)W.sub.S.sub.--.sub.ACT(m--
2Mmi)TRQ(m-Mmi)TRQ(m-2Mmi)] (65)
[0307] Herein, the notation "m" is a notation indicating
discretized time, and indicates data detected or calculated every
aforementioned control period .DELTA.Tc. In addition, this control
cycle .DELTA.Tc and model sampling period .DELTA.Tm satisfy the
following formula (66).
.DELTA.Tm=Mmc.DELTA.Tc (66)
[0308] According to the recursive identification algorithms, the
current control value .theta.'(m) of the model parameter vector
defining the model parameters a1', a2', b1' and b2' as components
is calculated from the previous control value .theta.'(m-1) by the
following formula (67).
.theta.'(m)=.theta.'(m-1)+KP'(m)ide'(m) (67)
[0309] Herein, ide'(m) in the above formula (67) indicates
identification error defined by the following formula (68). KP'(m)
indicates a gain coefficient vector defined by the following
formula (69). P'(m) in the following formula (69) is the
fourth-order square matrix defined by the following formula (70).
In addition, I in the following formula (70) indicates a
fourth-order unit matrix, and .lamda.1' and .lamda.2' respectively
indicate weighting parameters.
ide(m)=.theta.'.sup.T(m-1).zeta.(m) (68)
KP ' ( m ) = P ' ( m ) .zeta. ' ( m ) 1 + .zeta. ' ( m ) T P ' ( m
) .zeta. ' ( m ) ( 69 ) P ' ( m + 1 ) = 1 .lamda. 1 ' { I -
.lamda.2 ' P ' ( m ) .zeta. ' ( m ) .zeta. ' ( m ) T .lamda. 1 ' +
.lamda.2 ' .zeta. ' ( m ) T P ' ( m ) .zeta. ' ( m ) } P ' ( m ) (
70 ) ##EQU00020##
[0310] Herein, the various recursive identification algorithms can
be selected as shown below, by setting the weighting parameters
.lamda.1' and .lamda.2' of the above formula (70) as follows.
[0311] .lamda.1'=1, .lamda.2'=0: fixed gain algorithm
[0312] .lamda.1'=1, .lamda.2'=1: least-squares method algorithm
[0313] .lamda.1'=1, .lamda.2'=.lamda.: gradual decrease algorithm
(0<.lamda.<1)
[0314] .lamda.1'=.lamda., .lamda.2'=1: weighted least-squared
method algorithm (0<.lamda.<1)
[0315] By occasionally identifying the model parameters a1', a2',
b1' and b2' by way of the identifier 75 in the above way, it is
possible to prevent the control precision from declining, even in a
case of modeling error arising in the plant model described later
due to individual variation or degradation over time of the
vehicle.
Configuration of Feedback Controller
[0316] The configuration of the feedback controller 71 will be
explained.
[0317] The feedback controller 71 is configured to include the
aforementioned target value correction unit 742 that calculates the
corrected target wheel speed W.sub.S.sub.--.sub.CMD.sub.--.sub.MOD,
the predictor 712 that calculates the predicted drive wheel speed
PREW.sub.S.sub.--.sub.EXS as a future predicted value of the drive
wheel speed W.sub.S.sub.--.sub.ACT, a evaluation function value
calculating unit 713 that calculates an evaluation function value
J'' based on this corrected target wheel speed
W.sub.S.sub.--.sub.CMD.sub.--.sub.MOD and predicted drive wheel
speed PREW.sub.S.sub.--.sub.EXS, and the extremum searching
optimization unit 714 that calculates an optimum torque
DTRQ.sub.OPT and search input DTRQ.sub.EXS such that the evaluation
function value J'' calculated becomes an extremum, i.e. becomes a
minimum value.
[0318] In addition, the extremum searching optimization unit 714
searches for a minimum value of the evaluation function value J''
using the periodic reference signal S.sub.REF' as described in
detail later. Consequently, the extremum searching optimization
unit 714 inputs the optimum torque DTRQ.sub.OPT not containing the
component of this reference signal S.sub.REF and the search input
DTRQ.sub.EXS containing the component of the reference signal
S.sub.REF into the aforementioned predictor 712.
[0319] In addition, the optimum torque DTRQ.sub.OPT not containing
the component of the reference signal S.sub.REF' becomes the FB
torque DTRQ.sub.FB.
[0320] Moreover, the control cycle of this feedback controller 71
is basically .DELTA.Te, which is shorter than the update period
.DELTA.Tc of the target torque TRQ. Therefore, the update period
.DELTA.Te of this optimum torque DTRQ.sub.OPT and search input
DTRQ.sub.EXS becomes shorter than the update period .DELTA.Tc of
the target torque TRQ. The details of this control cycle .DELTA.Te
will be explained later in detail.
[0321] Hereinafter, the configurations of this predictor 712,
evaluation function value calculating unit 713, and extremum
searching optimization unit 714 will be explained in order.
Configuration of Predictor
[0322] The configuration of the predictor 712 will be
explained.
[0323] The predictor 712 calculates a predicted drive wheel speed
PREW.sub.S.sub.--.sub.EXS based on the drive wheel speed
W.sub.S.sub.--.sub.ACT, target torque TRQ, and search input
TRQ.sub.EXS containing the component of the reference signal
S.sub.REF'. Herein, the sum of adding, by way of the adder 77, the
FF torque TRQ.sub.FF to the search input DTRQ.sub.EXS calculated by
the extremum searching optimization unit 714 is used in the search
input TRQ.sub.EXS input to the predictor 712. In addition, when
calculating the predicted drive wheel speed
PREW.sub.S.sub.--.sub.EXS, this predictor 712 uses a model of the
vehicle (hereinafter referred to as "plant model") showing the
dynamic characteristic of the drive wheel speed
W.sub.S.sub.--.sub.ACT from the target torque TRQ.
[0324] The optimum form of this plant model is derived as the
following formulas (71) to (79), similarly to the formulas (27) to
(35) of the plant model in the aforementioned first embodiment.
[0325] Predicted drive wheel speed after 1--model sample time
PREW S_EXS ( n + MmcMce ) = a 1 ' ( n , k - 1 ) W S_ACT ( n ) + a 2
' ( n , k - 1 ) W S_ACT ( n - MmiMie ) + b 1 ' ( n , k - 1 ) TRQ
EXS ( n ) + b 2 ' ( n , k - 1 ) TRQ ( n , k - 1 ) ( .apprxeq. W
S_ACT ( k + 1 ) ) ( 71 ) ##EQU00021##
[0326] Predicted drive wheel speed after 2--model sample times
PREW S_EXS ( n + 2 MmcMce ) = a 1 ' ( n , k - 1 ) PREW S_CNT ( n +
MmcMce ) + a 2 ' ( n , k - 1 ) W S_ACT ( n ) + b 1 ' ( n , k - 1 )
TRQ EXS ( n ) + b 2 ' ( n , k - 1 ) TRQ ( n , k ) ( .apprxeq. W
S_ACT ( k + 2 ) ) ( 72 ) PREW S_CNT ( n + MmiMie ) = a 1 ' ( n , k
- 1 ) W S_ACT ( n ) + a 2 ' ( n , k - 1 ) W S_ACT ( n - MmiMie ) +
b 1 ' ( n , k - 1 ) TRQ ( n , k ) + b 2 ' ( n , k - 1 ) TRQ ( n , k
- 1 ) ( 73 ) ##EQU00022##
[0327] Predicted drive wheel speed after 3--model sample times
PREW S_EXS ( n + 3 MmcMce ) = a 1 ' ( n , k - 1 ) PREW S_CNT ( n +
2 MmcMce ) + a 2 ' ( n , k - 1 ) PREW S_CNT ( n + MmcMce ) + b 1 '
( n , k - 1 ) TRQ EXS ( n ) + b 2 ' ( n , k - 1 ) TRQ ( n , k ) (
.apprxeq. W S_ACT ( k + 3 ) ) ( 74 ) PREW S_CNT ( n + 2 MmcMce ) =
a 1 ' ( n , k - 1 ) PREW S_CNT ( n + MmcMce ) + a 2 ' ( n , k - 1 )
W S_ACT ( n ) + b 1 ' ( n , k - 1 ) TRQ ( n , k ) + b 2 ' ( n , k -
1 ) TRQ ( n , k ) ( 75 ) ##EQU00023##
[0328] Predicted drive wheel speed after 4--model sample times
PREW S_EXS ( n + 4 MmcMce ) = a 1 ' ( n , k - 1 ) PREW S_CNT ( n +
3 MmcMce ) + a 2 ' ( n , k - 1 ) PREW S_CNT ( n + 2 MmcMce ) + b 1
' ( n , k - 1 ) TRQ EXS ( n ) + b 2 ' ( n , k - 1 ) TRQ ( n , k ) (
.apprxeq. W S_ACT ( k + 4 ) ) ( 76 ) PREW S_CNT ( n + 3 MmcMce ) =
a 1 ' ( n , k - 1 ) PREW S_CNT ( n + 2 MmcMce ) + a 2 ' ( n , k - 1
) PREW S_CNT ( n + MmcMce ) + b 1 ' ( n , k - 1 ) TRQ ( n , k ) + b
2 ' ( n , k - 1 ) TRQ ( n , k ) ( 77 ) ##EQU00024##
[0329] Predicted drive wheel speed after mp--model sample times
PREW S_EXS ( n + mpMmcMce ) = a 1 ' ( n , k - 1 ) PREW S_CNT ( n +
( mp - 1 ) MmcMce ) + a 2 ' ( n , k - 1 ) PREW S_CNT ( n + ( mp - 2
) MmcMce ) + b 1 ' ( n , k - 1 ) TRQ EXS ( n ) + b 2 ' ( n , k - 1
) TRQ ( n , k ) ( .apprxeq. W S_ACT ( k + mp ) ) ( 78 ) PREW S_CNT
( n + ( mp - 1 ) MmcMce ) = a 1 ' ( n , k - 1 ) PREW S_CNT ( n + (
mp - 2 ) MmcMce ) + a 2 ' ( n , k - 1 ) PREW S_CNT ( n + ( mp - 1 )
MmcMce ) + b 1 ' ( n , k - 1 ) TRQ ( n , k ) + b 2 ' ( n , k - 1 )
TRQ ( n , k ) ( 79 ) ##EQU00025##
[0330] Herein, the notation "n" is a notation indicating
discretized time, and indicates being data detected or calculated
every control cycle .DELTA.Te. In addition, this control cycle
.DELTA.Te and model sampling period .DELTA.Tm satisfy the following
formula (80).
.DELTA.Tm=Mmc.DELTA.Tc=Mmc Mce.DELTA.Te (80)
[0331] Moreover, in the above formulas (71) to (79), the model
parameters a1'(n,k), a2'(n,k), b1'(n,k) and b2'(n,k) and the target
torque TRQ(n,k) respectively indicate parameters updated at the
period .DELTA.Tc as described above, and being parameters
oversampled at the shorter period .DELTA.Te.
Configuration of evaluation Function Value Calculating Unit
[0332] The configuration of the evaluation function value
calculating unit 713 will be explained.
[0333] The evaluation function value calculating unit 713
calculates the evaluation function value J'', such as that shown in
the following formula (81), based on the square sum of the
deviation between the predicted drive wheel speed
PREW.sub.S.sub.--.sub.EXS calculated by the predictor 712, and the
corrected target wheel speed W.sub.S.sub.--.sub.CMD.sub.--.sub.MOD
calculated by the target value correction unit 742.
J '' ( n ) = i = 1 mp ( PREW S_EXS ( n + MmcMcei ) - W S_CMD _MOD (
k + 1 ) ) 2 ( 81 ) ##EQU00026##
Configuration of Extremum Searching Optimization Unit
[0334] The configuration of the extremum searching optimization
unit 714 will be explained.
[0335] The extremum searching optimization unit 714 calculates the
optimum torque DTRQ.sub.OPT and search input DTRQ.sub.EXS such that
the evaluation function value J'' becomes a minimum by the sequence
shown in the following formulas (82) to (92). More specifically, it
first calculates the moving average value C.sub.R.sub.--.sub.AVE'
using the reference signal S.sub.REF', based on the following
formulas (82) to (87).
J*(n)=-J''(n) (82)
J.sub.WF'(n)=0.5J*(n)-0.5J*(n-1) (83)
S.sub.REF'(n)=A.sub.REF sin(2.pi.F.sub.REF'n.DELTA.Te) (84)
C.sub.R'(n)=J.sub.WF'(n)S.sub.REF'(n) (85)
C R_AVE ' ( n ) = i = 0 N AVE ' C R ' ( n - i ) ( 86 ) N AVE ' = j
1 F REF ' .DELTA. Te ( 87 ) ##EQU00027##
[0336] Furthermore, it calculates the optimum torque DTRQ.sub.OPT
and search input DTRQ.sub.EXS based on the algorithm of sliding
mode control as shown in the following formulas (88) to (92), so
that the moving average value C.sub.R.sub.--.sub.AVE' becomes
"0".
DTRQ.sub.OPT(n)=DTRQ.sub.RCH(n)+DTRQ.sub.ADP(n) (88)
DTRQ.sub.RCH(n)=K.sub.RCH'.sigma.'(n) (89)
DTRQ ADP ( n ) = K ADP ' i = 1 n .sigma. ' ( i ) ( 90 )
##EQU00028##
.sigma.'(n)=C.sub.R.sub.--.sub.AVE'(n)+S'C.sub.R.sub.--.sub.AVE'(n-1)
(91)
DTRQ.sub.EXS(n)=DTRQ.sub.OPT(n)+S.sub.REF'(n) (92)
Configuration of Feedforward Controller
[0337] The configuration of the feedforward controller 72 will be
explained.
[0338] The feedforward controller 72 calculates the FF torque
TRQ.sub.FF based on the following formula (93).
TRQ.sub.FF(m)=KTRQ.sub.MIN(k)+KTRQ(k)(KTRQ.sub.MAX(k)-KTRQ.sub.MIN(k))
(93)
[0339] In the above formula (93), the torque values KTRQ.sub.MIN
and KTRQ.sub.MAX are determined according to the engine revolution
sped NE, based on the map shown in FIG. 13. In addition, the
coefficient KTRQ is determined according to the accelerator opening
AP, based on the map shown in FIG. 14.
[0340] Next, the results of an example of control by the ECU of the
present embodiment will be explained while referring to FIG.
15.
[0341] FIG. 15 shows examples of control, in which FIG. 15(a) shows
an example of control by a conventional PID controller, and FIG.
15(b) shows an example of control by the ECU of the present
embodiment. It should be noted that FIG. 15 shows examples of
control in a case in which the accelerator is depressed at the time
t.sub.E from a state in which the vehicle is stopped, and
thereafter, the road surface coefficient of friction suddenly
declines at the time t.sub.F.
[0342] As shown in FIG. 15(a), in the case of using convention PID
control, if the accelerator is depressed and the target wheel speed
W.sub.S.sub.--.sub.CMD changes step-wise, the drive wheel speed
W.sub.S.sub.--.sub.ACT shows overshoot and oscillatory behavior
relative to this target wheel speed W.sub.S.sub.--.sub.CMD. In
other words, both the acceleration and the stability of the vehicle
decline considerably. In contrast, as shown in FIG. 15(b), such
overshoot and oscillatory behavior is suppressed in a case of using
the ECU of the present embodiment. Therefore, both the acceleration
and stability of the vehicle can be maintained.
[0343] In addition, even in a case of the road surface coefficient
of friction declining, the drive wheel speed W.sub.S.sub.--.sub.ACT
oscillates greatly relative to the target wheel speed
W.sub.S.sub.--.sub.CMD, and the behavior of the vehicle is
unsettled in the case of using conventional PID control. In
addition, excessive deceleration occurs accompanying this. In
contrast, in the case of using the ECU of the present embodiment,
stable driving is made possible, without the behavior of the
vehicle being unsettled in this way.
[0344] It should be noted that the present invention is not limited
to the aforementioned embodiments, and various modifications
thereto are possible.
[0345] Although the aforementioned first embodiment and second
embodiment have explained a control device of an exhaust
purification device and a traction control system as plants having
a characteristic of a large response delay, it is not limited to
this. The present invention is effective for any type of plant, so
long as being a control object having a large response delay. In
particular, it is more effective if a plant in which the dead time
is short relative to the length of the response delay.
[0346] For the ECU 3 of the above-mentioned first embodiment,
although the exhaust purification device 2 equipped with the two
selective reduction catalysts 231 and 232 is set as the control
object, it is not limited thereto. The controller of the present
invention can quickly and highly precisely control a detected value
in a plant having a characteristic of a large response delay to a
target value. Therefore, even in a case of an exhaust purification
device equipped with only one selective reduction catalyst being
set as the control object, it is possible to suppress ammonia slip
while maintaining a high NOx reduction rate.
[0347] In addition, for the control device of the afore-mentioned
second embodiment, although a vehicle equipped with an engine as a
drive source is set as the control object, it is not limited
thereto. For example, it may be a fuel cell vehicle or electric
automobile with a motor as the drive source.
* * * * *